sweeps.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "eb3d4a87",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "from glob import glob\n",
    "import numpy as np\n",
    "import re\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "ecac06bb",
   "metadata": {},
   "outputs": [],
   "source": [
    "def getSweepData(data_folders, param):\n",
    "    \n",
    "    scores    = pd.DataFrame()\n",
    "    wers_dict = dict()\n",
    "    for sweep in data_folders:\n",
    "\n",
    "        with open(os.path.join(sweep, 'log.txt')) as f:\n",
    "            lines = f.readlines()\n",
    "\n",
    "        log  = ' '.join(np.asarray(lines))\n",
    "        try:\n",
    "            val    = float(re.findall(fr'{param}=(\\d+.\\d+)', log)[0])\n",
    "            wer    = float(re.findall(r'validation WER: (\\d+.\\d+)', log)[-1])\n",
    "            scores = scores.append([{param : val, \n",
    "                                     'WER' : wer,\n",
    "                                     }])\n",
    "            wers_dict[val] = np.asarray([float(x) for x in re.findall(r'validation WER: (\\d+.\\d+)', log)])\n",
    "        except:\n",
    "            continue\n",
    "            \n",
    "    return scores, wers_dict\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "id": "4613c8d4",
   "metadata": {},
   "outputs": [],
   "source": [
    "import neptune"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "id": "51a17482",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<module 'neptune.new' from '/oak/stanford/groups/shenoy/ghwilson/miniconda3/envs/mag-3.9/lib/python3.9/site-packages/neptune/new/__init__.py'>"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "neptune.new.ini"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "10943b69",
   "metadata": {},
   "outputs": [],
   "source": [
    "A= np.ones((1, 80, 6000))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "id": "92058b68",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1, 80, 5000)"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "A[...,:5000].shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "908311f5",
   "metadata": {},
   "source": [
    "### HP sweep\n",
    "\n",
    "Let's look at performance as a function of hyperparams"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c0403ac7",
   "metadata": {},
   "source": [
    "best thus far, 34% WER:\n",
    " - 1000 Hz data, seqlen = 600 (* 8)\n",
    " - model_size = 1300\n",
    " - num_layers = 8\n",
    " - dropout = 0.2\n",
    " - learning_rate = 1e-3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "83df217a",
   "metadata": {},
   "outputs": [],
   "source": [
    "data_dir = '/oak/stanford/projects/babelfish/magneto/GaddyPaper/output/efficiency_sweeps/GaddyTransformer/1/log.txt'\n",
    "with open(data_dir) as f:\n",
    "    lines = f.readlines()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "9c1ad6a4",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['3da3532cb8ba63fa524c0e776c845a19cd61b7a9\\n',\n",
       " '\\n',\n",
       " 'diff --git a/architecture.py b/architecture.py\\n',\n",
       " 'index 4fc3793..771be7e 100644\\n',\n",
       " '--- a/architecture.py\\n',\n",
       " '+++ b/architecture.py\\n',\n",
       " '@@ -1,11 +1,14 @@\\n',\n",
       " ' import random\\n',\n",
       " '-\\n',\n",
       " ' import torch\\n',\n",
       " ' from torch import nn\\n',\n",
       " ' import torch.nn.functional as F\\n',\n",
       " '-\\n',\n",
       " ' from transformer import TransformerEncoderLayer\\n',\n",
       " ' \\n',\n",
       " '+\\n',\n",
       " '+import sys\\n',\n",
       " \"+sys.path.append('/home/users/ghwilson/projects/subvocal_speech/magneto/models/')\\n\",\n",
       " '+from s4d import S4D\\n',\n",
       " '+\\n',\n",
       " ' from absl import flags\\n',\n",
       " ' FLAGS = flags.FLAGS\\n',\n",
       " \" flags.DEFINE_integer('model_size', 768, 'number of hidden dimensions')\\n\",\n",
       " '@@ -39,7 +42,8 @@ class ResBlock(nn.Module):\\n',\n",
       " '             res = input_value\\n',\n",
       " ' \\n',\n",
       " '         return F.relu(x + res)\\n',\n",
       " '-\\n',\n",
       " '+    \\n',\n",
       " '+    \\n',\n",
       " ' class Model(nn.Module):\\n',\n",
       " '     def __init__(self, num_features, num_outs, num_aux_outs=None):\\n',\n",
       " '         super().__init__()\\n',\n",
       " '@@ -50,10 +54,10 @@ class Model(nn.Module):\\n',\n",
       " '             ResBlock(FLAGS.model_size, FLAGS.model_size, 2),\\n',\n",
       " '         )\\n',\n",
       " '         self.w_raw_in = nn.Linear(FLAGS.model_size, FLAGS.model_size)\\n',\n",
       " '-\\n',\n",
       " '-        encoder_layer = TransformerEncoderLayer(d_model=FLAGS.model_size, nhead=8, relative_positional=True, relative_positional_distance=100, dim_feedforward=3072, dropout=FLAGS.dropout)\\n',\n",
       " '+        encoder_layer = TransformerEncoderLayer(d_model=FLAGS.model_size, nhead=8, relative_positional=True, \\n',\n",
       " '+                                                relative_positional_distance=100, dim_feedforward=3072, dropout=FLAGS.dropout)\\n',\n",
       " '         self.transformer = nn.TransformerEncoder(encoder_layer, FLAGS.num_layers)\\n',\n",
       " '-        self.w_out = nn.Linear(FLAGS.model_size, num_outs)\\n',\n",
       " '+        self.w_out       = nn.Linear(FLAGS.model_size, num_outs)\\n',\n",
       " ' \\n',\n",
       " '         self.has_aux_out = num_aux_outs is not None\\n',\n",
       " '         if self.has_aux_out:\\n',\n",
       " '@@ -67,14 +71,13 @@ class Model(nn.Module):\\n',\n",
       " '             if r > 0:\\n',\n",
       " '                 x_raw[:,:-r,:] = x_raw[:,r:,:] # shift left r\\n',\n",
       " '                 x_raw[:,-r:,:] = 0\\n',\n",
       " '-\\n',\n",
       " '+        \\n',\n",
       " '         x_raw = x_raw.transpose(1,2) # put channel before time for conv\\n',\n",
       " '         x_raw = self.conv_blocks(x_raw)\\n',\n",
       " '         x_raw = x_raw.transpose(1,2)\\n',\n",
       " '         x_raw = self.w_raw_in(x_raw)\\n',\n",
       " ' \\n',\n",
       " '         x = x_raw\\n',\n",
       " '-\\n',\n",
       " '         x = x.transpose(0,1) # put time first\\n',\n",
       " '         x = self.transformer(x)\\n',\n",
       " '         x = x.transpose(0,1)\\n',\n",
       " '@@ -83,4 +86,144 @@ class Model(nn.Module):\\n',\n",
       " '             return self.w_out(x), self.w_aux(x)\\n',\n",
       " '         else:\\n',\n",
       " '             return self.w_out(x)\\n',\n",
       " '+        \\n',\n",
       " ' \\n',\n",
       " '+    \\n',\n",
       " '+class S4Model(nn.Module):\\n',\n",
       " '+    def __init__(self, num_features, num_outs, num_aux_outs=None):\\n',\n",
       " '+        super().__init__()\\n',\n",
       " '+        self.prenorm = False \\n',\n",
       " '+        \\n',\n",
       " '+        # Linear encoder\\n',\n",
       " '+        self.encoder = nn.Linear(8, FLAGS.model_size)\\n',\n",
       " '+        \\n',\n",
       " '+        # Stack S4 layers as residual blocks\\n',\n",
       " '+        self.s4_layers = nn.ModuleList()\\n',\n",
       " '+        self.norms     = nn.ModuleList()\\n',\n",
       " '+        self.dropouts  = nn.ModuleList()\\n',\n",
       " '+        for _ in range(FLAGS.num_layers):\\n',\n",
       " '+            self.s4_layers.append(\\n',\n",
       " '+                S4D(FLAGS.model_size, dropout=FLAGS.dropout, transposed=True,\\n',\n",
       " '+                    lr=0.001)\\n',\n",
       " '+            )\\n',\n",
       " '+            self.norms.append(nn.LayerNorm(FLAGS.model_size))\\n',\n",
       " '+            self.dropouts.append(nn.Dropout1d(FLAGS.dropout))\\n',\n",
       " '+\\n',\n",
       " '+        \\n',\n",
       " '+        self.w_out = nn.Linear(FLAGS.model_size, num_outs)\\n',\n",
       " '+        \\n',\n",
       " '+        self.has_aux_out = num_aux_outs is not None\\n',\n",
       " '+        if self.has_aux_out:\\n',\n",
       " '+            self.w_aux = nn.Linear(FLAGS.model_size, num_aux_outs)\\n',\n",
       " '+\\n',\n",
       " '+            \\n',\n",
       " '+    def forward(self, x_feat, x_raw, session_ids):\\n',\n",
       " '+        # x shape is (batch, time, electrode)\\n',\n",
       " '+\\n',\n",
       " '+        if self.training:\\n',\n",
       " '+            r = random.randrange(8)\\n',\n",
       " '+            if r > 0:\\n',\n",
       " '+                x_raw[:,:-r,:] = x_raw[:,r:,:] # shift left r\\n',\n",
       " '+                x_raw[:,-r:,:] = 0\\n',\n",
       " '+                \\n',\n",
       " '+        x = self.encoder(x_raw) \\n',\n",
       " '+        x = x.transpose(-1, -2)  # (B, L, d_model) -> (B, d_model, L)\\n',\n",
       " '+        for i, (layer, norm, dropout) in enumerate(zip(self.s4_layers, self.norms, self.dropouts)):\\n',\n",
       " '+            # Each iteration of this loop will map (B, d_model, L) -> (B, d_model, L)\\n',\n",
       " '+\\n',\n",
       " '+            z = x\\n',\n",
       " '+            if self.prenorm:\\n',\n",
       " '+                # Prenorm\\n',\n",
       " '+                z = norm(z.transpose(-1, -2)).transpose(-1, -2)\\n',\n",
       " '+\\n',\n",
       " '+            # Apply S4 block: we ignore the state input and output\\n',\n",
       " '+            z, _ = layer(z)\\n',\n",
       " '+\\n',\n",
       " '+            # Dropout on the output of the S4 block\\n',\n",
       " '+            z = dropout(z)\\n',\n",
       " '+\\n',\n",
       " '+            # Residual connection\\n',\n",
       " '+            x = z + x\\n',\n",
       " '+        \\n',\n",
       " '+            if not self.prenorm:\\n',\n",
       " '+                # Postnorm\\n',\n",
       " '+                x = norm(x.transpose(-1, -2)).transpose(-1, -2)\\n',\n",
       " '+            \\n',\n",
       " '+            if i < 3:\\n',\n",
       " '+                x = x[:, :, ::2]\\n',\n",
       " '+\\n',\n",
       " '+        #x = x[:, :, ::8]\\n',\n",
       " '+        x = x.transpose(-1, -2)\\n',\n",
       " '+\\n',\n",
       " '+        if self.has_aux_out:\\n',\n",
       " '+            return self.w_out(x), self.w_aux(x)\\n',\n",
       " '+        else:\\n',\n",
       " '+            return self.w_out(x)\\n',\n",
       " '+        \\n',\n",
       " '+        \\n',\n",
       " '+\\n',\n",
       " \"+'''\\n\",\n",
       " '+class S4model(WavBase):\\n',\n",
       " '+    \"\"\"\\n',\n",
       " '+    https://github.com/HazyResearch/state-spaces/blob/ab287c63f4938a76d06a6b6868ee4a7163b50b05/example.py\\n',\n",
       " '+    \"\"\"\\n',\n",
       " '+    def __init__(self, params: BaseParams):\\n',\n",
       " '+        super().__init__(params)\\n',\n",
       " '+        # Linear encoder\\n',\n",
       " '+        self.encoder = nn.Linear(params.in_channels, params.d_model)\\n',\n",
       " '+        \\n',\n",
       " '+        # Stack S4 layers as residual blocks\\n',\n",
       " '+        self.s4_layers = nn.ModuleList()\\n',\n",
       " '+        self.norms = nn.ModuleList()\\n',\n",
       " '+        self.dropouts = nn.ModuleList()\\n',\n",
       " '+        for _ in range(params.n_layers):\\n',\n",
       " '+            self.s4_layers.append(\\n',\n",
       " '+                S4D(params.d_model, dropout=params.dropout, transposed=True,\\n',\n",
       " '+                    lr=params.lr)\\n',\n",
       " '+            )\\n',\n",
       " '+            self.norms.append(nn.LayerNorm(params.d_model))\\n',\n",
       " '+            self.dropouts.append(nn.Dropout1d(params.dropout))\\n',\n",
       " '+\\n',\n",
       " '+        # Linear decoder\\n',\n",
       " '+        self.decoder = nn.Linear(params.d_model, params.num_words)\\n',\n",
       " '+        \\n',\n",
       " '+        self.prenorm = params.prenorm\\n',\n",
       " '+        \\n',\n",
       " '+    def forward(self, x, extra=None):\\n',\n",
       " '+        \"\"\"\\n',\n",
       " '+        Input x is shape (B, L, d_input)\\n',\n",
       " '+        \"\"\"\\n',\n",
       " '+        x = self.encoder(x)  # (B, L, d_input) -> (B, L, d_model)\\n',\n",
       " '+\\n',\n",
       " '+        x = x.transpose(-1, -2)  # (B, L, d_model) -> (B, d_model, L)\\n',\n",
       " '+        for layer, norm, dropout in zip(self.s4_layers, self.norms, self.dropouts):\\n',\n",
       " '+            # Each iteration of this loop will map (B, d_model, L) -> (B, d_model, L)\\n',\n",
       " '+\\n',\n",
       " '+            z = x\\n',\n",
       " '+            if self.params.prenorm:\\n',\n",
       " '+                # Prenorm\\n',\n",
       " '+                z = norm(z.transpose(-1, -2)).transpose(-1, -2)\\n',\n",
       " '+\\n',\n",
       " '+            # Apply S4 block: we ignore the state input and output\\n',\n",
       " '+            z, _ = layer(z)\\n',\n",
       " '+\\n',\n",
       " '+            # Dropout on the output of the S4 block\\n',\n",
       " '+            z = dropout(z)\\n',\n",
       " '+\\n',\n",
       " '+            # Residual connection\\n',\n",
       " '+            x = z + x\\n',\n",
       " '+\\n',\n",
       " '+            if not self.params.prenorm:\\n',\n",
       " '+                # Postnorm\\n',\n",
       " '+                x = norm(x.transpose(-1, -2)).transpose(-1, -2)\\n',\n",
       " '+\\n',\n",
       " '+        x = x.transpose(-1, -2)\\n',\n",
       " '+\\n',\n",
       " '+        # Pooling: average pooling over the sequence length\\n',\n",
       " '+        x = x.mean(dim=1)\\n',\n",
       " '+\\n',\n",
       " '+        # Decode the outputs\\n',\n",
       " '+        x = self.decoder(x)  # (B, d_model) -> (B, d_output)\\n',\n",
       " '+        x = F.log_softmax(x, dim=1)\\n',\n",
       " '+        return x\\n',\n",
       " \"+'''\\n\",\n",
       " 'diff --git a/asr_evaluation.py b/asr_evaluation.py\\n',\n",
       " 'index 7c3083e..be441d7 100644\\n',\n",
       " '--- a/asr_evaluation.py\\n',\n",
       " '+++ b/asr_evaluation.py\\n',\n",
       " '@@ -1,31 +1,56 @@\\n',\n",
       " ' import os\\n',\n",
       " ' import logging\\n',\n",
       " ' \\n',\n",
       " '-import deepspeech\\n',\n",
       " '+#import deepspeech\\n',\n",
       " ' import jiwer\\n',\n",
       " ' import soundfile as sf\\n',\n",
       " ' import numpy as np\\n',\n",
       " ' from unidecode import unidecode\\n',\n",
       " ' import librosa\\n',\n",
       " '+import whisper\\n',\n",
       " '+\\n',\n",
       " '+\\n',\n",
       " '+removeBlanks = True\\n',\n",
       " '+\\n',\n",
       " ' \\n',\n",
       " ' def evaluate(testset, audio_directory):\\n',\n",
       " \"-    model = deepspeech.Model('deepspeech-0.7.0-models.pbmm')\\n\",\n",
       " \"-    model.enableExternalScorer('deepspeech-0.7.0-models.scorer')\\n\",\n",
       " '+    \\n',\n",
       " \"+    print('Loading Whisper ASR model...')\\n\",\n",
       " '+    model = whisper.load_model(\"small.en\")\\n',\n",
       " \"+    print('Done! Transcribing audio files.')\\n\",\n",
       " \"+    #model = deepspeech.Model('deepspeech-0.7.0-models.pbmm')\\n\",\n",
       " \"+    #model.enableExternalScorer('deepspeech-0.7.0-models.scorer')\\n\",\n",
       " '     predictions = []\\n',\n",
       " '     targets = []\\n',\n",
       " '     for i, datapoint in enumerate(testset):\\n',\n",
       " \"-        audio, rate = sf.read(os.path.join(audio_directory,f'example_output_{i}.wav'))\\n\",\n",
       " '-        if rate != 16000:\\n',\n",
       " '-            audio = librosa.resample(audio, rate, 16000)\\n',\n",
       " \"-        assert model.sampleRate() == 16000, 'wrong sample rate'\\n\",\n",
       " '-        audio_int16 = (audio*(2**15)).astype(np.int16)\\n',\n",
       " '-        text = model.stt(audio_int16)\\n',\n",
       " '-        predictions.append(text)\\n',\n",
       " \"-        target_text = unidecode(datapoint['text'])\\n\",\n",
       " '+        \\n',\n",
       " \"+        text = model.transcribe(os.path.join(audio_directory,f'example_output_{i}.wav'))\\n\",\n",
       " \"+        predictions.append(text['text'])\\n\",\n",
       " \"+        target_text = unidecode(str(datapoint['text'][0]))\\n\",\n",
       " '         targets.append(target_text)\\n',\n",
       " '+\\n',\n",
       " '+        # old deepspeech code:\\n',\n",
       " \"+        #audio, rate = sf.read(os.path.join(audio_directory,f'example_output_{i}.wav'))\\n\",\n",
       " '+        #if rate != 16000:\\n',\n",
       " '+        #    audio = librosa.resample(audio, rate, 16000)\\n',\n",
       " \"+        #assert model.sampleRate() == 16000, 'wrong sample rate'\\n\",\n",
       " '+        #audio_int16 = (audio*(2**15)).astype(np.int16)\\n',\n",
       " '+        #text = model.stt(audio_int16)\\n',\n",
       " '+        #predictions.append(text)\\n',\n",
       " \"+        #target_text = unidecode(datapoint['text'])\\n\",\n",
       " '+        #targets.append(target_text)\\n',\n",
       " '+\\n',\n",
       " '+        \\n',\n",
       " '     transformation = jiwer.Compose([jiwer.RemovePunctuation(), jiwer.ToLowerCase()])\\n',\n",
       " '-    targets = transformation(targets)\\n',\n",
       " '+    targets     = transformation(targets)\\n',\n",
       " '     predictions = transformation(predictions)\\n',\n",
       " '+    \\n',\n",
       " '+    if removeBlanks:\\n',\n",
       " '+        for i, (targ, pred) in enumerate(zip(targets, predictions)):\\n',\n",
       " \"+            if targ == '':\\n\",\n",
       " '+                del targets[i]\\n',\n",
       " '+                del predictions[i]\\n',\n",
       " '+    \\n',\n",
       " \"     logging.info(f'targets: {targets}')\\n\",\n",
       " \"     logging.info(f'predictions: {predictions}')\\n\",\n",
       " \"     logging.info(f'wer: {jiwer.wer(targets, predictions)}')\\n\",\n",
       " 'diff --git a/data_utils.py b/data_utils.py\\n',\n",
       " 'index fe3a10e..6274b89 100644\\n',\n",
       " '--- a/data_utils.py\\n',\n",
       " '+++ b/data_utils.py\\n',\n",
       " '@@ -62,7 +62,7 @@ def mel_spectrogram(y, n_fft, num_mels, sampling_rate, hop_size, win_size, fmin,\\n',\n",
       " '     return spec\\n',\n",
       " ' \\n',\n",
       " ' def load_audio(filename, start=None, end=None, max_frames=None, renormalize_volume=False):\\n',\n",
       " '-    audio, r = sf.read(filename)\\n',\n",
       " \"+    audio, r = sf.read(filename.replace('flac', 'wav'))\\n\",\n",
       " ' \\n',\n",
       " '     if len(audio.shape) > 1:\\n',\n",
       " '         audio = audio[:,0] # select first channel of stero audio\\n',\n",
       " 'diff --git a/evaluate.py b/evaluate.py\\n',\n",
       " 'index e1fbb8f..4419342 100644\\n',\n",
       " '--- a/evaluate.py\\n',\n",
       " '+++ b/evaluate.py\\n',\n",
       " '@@ -7,7 +7,7 @@ from torch import nn\\n',\n",
       " ' \\n',\n",
       " ' from architecture import Model\\n',\n",
       " ' from transduction_model import test, save_output\\n',\n",
       " '-from read_emg import EMGDataset\\n',\n",
       " '+from read_emg import EMGDataset, PreprocessedEMGDataset\\n',\n",
       " ' from asr_evaluation import evaluate\\n',\n",
       " ' from data_utils import phoneme_inventory, print_confusion\\n',\n",
       " ' from vocoder import Vocoder\\n',\n",
       " '@@ -16,6 +16,7 @@ from absl import flags\\n',\n",
       " ' FLAGS = flags.FLAGS\\n',\n",
       " \" flags.DEFINE_list('models', [], 'identifiers of models to evaluate')\\n\",\n",
       " \" flags.DEFINE_boolean('dev', False, 'evaluate dev insead of test')\\n\",\n",
       " \"+#flags.DEFINE_string('base_dir', '/oak/stanford/projects/babelfish/magneto/GaddyPaper/processed_data/', 'path to processed EMG dataset')\\n\",\n",
       " ' \\n',\n",
       " ' class EnsembleModel(nn.Module):\\n',\n",
       " '     def __init__(self, models):\\n',\n",
       " '@@ -39,7 +40,9 @@ def main():\\n',\n",
       " '             ], level=logging.INFO, format=\"%(message)s\")\\n',\n",
       " ' \\n',\n",
       " '     dev = FLAGS.dev\\n',\n",
       " '-    testset = EMGDataset(dev=dev, test=not dev)\\n',\n",
       " '+    #testset = EMGDataset(dev=dev, test=not dev)\\n',\n",
       " '+    testset   = PreprocessedEMGDataset(base_dir = FLAGS.base_dir, train = False, dev = FLAGS.dev, test = not FLAGS.dev)\\n',\n",
       " '+\\n',\n",
       " ' \\n',\n",
       " \"     device = 'cuda' if torch.cuda.is_available() else 'cpu'\\n\",\n",
       " ' \\n',\n",
       " 'diff --git a/read_emg.py b/read_emg.py\\n',\n",
       " 'index 664aa94..ca4eb6e 100644\\n',\n",
       " '--- a/read_emg.py\\n',\n",
       " '+++ b/read_emg.py\\n',\n",
       " '@@ -4,9 +4,10 @@ import numpy as np\\n',\n",
       " ' import matplotlib.pyplot as plt\\n',\n",
       " ' import random\\n',\n",
       " ' from collections import defaultdict\\n',\n",
       " '-import scipy\\n',\n",
       " '+import scipy.signal\\n',\n",
       " ' import json\\n',\n",
       " ' import copy\\n',\n",
       " '+import glob\\n',\n",
       " ' import sys\\n',\n",
       " ' import pickle\\n',\n",
       " ' import string\\n',\n",
       " '@@ -16,18 +17,25 @@ from copy import copy\\n',\n",
       " ' \\n',\n",
       " ' import librosa\\n',\n",
       " ' import soundfile as sf\\n',\n",
       " '+from scipy.io import loadmat\\n',\n",
       " ' \\n',\n",
       " '-import torch\\n',\n",
       " ' \\n',\n",
       " '+import torch\\n',\n",
       " ' from data_utils import load_audio, get_emg_features, FeatureNormalizer, phoneme_inventory, read_phonemes, TextTransform\\n',\n",
       " ' \\n',\n",
       " ' from absl import flags\\n',\n",
       " '+\\n',\n",
       " \"+data_folder    = '/oak/stanford/projects/babelfish/magneto/GaddyPaper/'\\n\",\n",
       " \"+project_folder = '/home/users/ghwilson/projects/silent_speech/'\\n\",\n",
       " '+\\n',\n",
       " '+\\n',\n",
       " ' FLAGS = flags.FLAGS\\n',\n",
       " \" flags.DEFINE_list('remove_channels', [], 'channels to remove')\\n\",\n",
       " \"-flags.DEFINE_list('silent_data_directories', ['./emg_data/silent_parallel_data'], 'silent data locations')\\n\",\n",
       " \"-flags.DEFINE_list('voiced_data_directories', ['./emg_data/voiced_parallel_data','./emg_data/nonparallel_data'], 'voiced data locations')\\n\",\n",
       " \"-flags.DEFINE_string('testset_file', 'testset_largedev.json', 'file with testset indices')\\n\",\n",
       " \"-flags.DEFINE_string('text_align_directory', 'text_alignments', 'directory with alignment files')\\n\",\n",
       " \"+flags.DEFINE_list('silent_data_directories', [f'{data_folder}/emg_data/silent_parallel_data'], 'silent data locations')\\n\",\n",
       " \"+flags.DEFINE_list('voiced_data_directories', [f'{data_folder}/emg_data/voiced_parallel_data',\\n\",\n",
       " \"+                                              f'{data_folder}/emg_data/nonparallel_data'], 'voiced data locations')\\n\",\n",
       " \"+flags.DEFINE_string('testset_file', f'{project_folder}/testset_largedev.json', 'file with testset indices')\\n\",\n",
       " \"+flags.DEFINE_string('text_align_directory', f'{data_folder}/text_alignments', 'directory with alignment files')\\n\",\n",
       " ' \\n',\n",
       " ' def remove_drift(signal, fs):\\n',\n",
       " \"     b, a = scipy.signal.butter(3, 2, 'highpass', fs=fs)\\n\",\n",
       " '@@ -143,6 +151,39 @@ class SizeAwareSampler(torch.utils.data.Sampler):\\n',\n",
       " '             batch.append(idx)\\n',\n",
       " '             batch_length += length\\n',\n",
       " '         # dropping last incomplete batch\\n',\n",
       " '+        \\n',\n",
       " '+class PreprocessedSizeAwareSampler(torch.utils.data.Sampler):\\n',\n",
       " '+    def __init__(self, emg_dataset, max_len):\\n',\n",
       " '+        self.dataset = emg_dataset\\n',\n",
       " '+        self.max_len = max_len\\n',\n",
       " '+\\n',\n",
       " '+    def __iter__(self):\\n',\n",
       " '+        indices = list(range(len(self.dataset)))\\n',\n",
       " '+        random.shuffle(indices)\\n',\n",
       " '+        batch = []\\n',\n",
       " '+        batch_length = 0\\n',\n",
       " '+        for idx in indices:\\n',\n",
       " '+           # directory_info, file_idx = self.dataset.example_indices[idx]\\n',\n",
       " '+            \\n',\n",
       " \"+            x         = loadmat(self.dataset.example_indices[idx])['audio_file'][0]\\n\",\n",
       " \"+            json_file = x.split('_audio_clean')[0] + '_info.json'\\n\",\n",
       " '+\\n',\n",
       " '+            with open(json_file) as f:\\n',\n",
       " '+                info = json.load(f)\\n',\n",
       " '+                \\n',\n",
       " '+                \\n',\n",
       " \"+            if not np.any([l in string.ascii_letters for l in info['text']]):\\n\",\n",
       " '+                continue\\n',\n",
       " \"+            length = sum([emg_len for emg_len, _, _ in info['chunks']])\\n\",\n",
       " '+            if length > self.max_len:\\n',\n",
       " \"+                logging.warning(f'Warning: example {idx} cannot fit within desired batch length')\\n\",\n",
       " '+            if length + batch_length > self.max_len:\\n',\n",
       " '+                yield batch\\n',\n",
       " '+                batch = []\\n',\n",
       " '+                batch_length = 0\\n',\n",
       " '+            batch.append(idx)\\n',\n",
       " '+            batch_length += length\\n',\n",
       " '+        # dropping last incomplete batch\\n',\n",
       " ' \\n',\n",
       " ' class EMGDataset(torch.utils.data.Dataset):\\n',\n",
       " '     def __init__(self, base_dir=None, limit_length=False, dev=False, test=False, no_testset=False, no_normalizers=False):\\n',\n",
       " '@@ -299,6 +340,118 @@ class EMGDataset(torch.utils.data.Dataset):\\n',\n",
       " \"                   'text_int':text_ints,\\n\",\n",
       " \"                   'text_int_lengths':text_lengths}\\n\",\n",
       " '         return result\\n',\n",
       " '+    \\n',\n",
       " '+    \\n',\n",
       " '+class PreprocessedEMGDataset(torch.utils.data.Dataset):\\n',\n",
       " '+    def __init__(self, base_dir=None, train = False, dev=False, test=False, limit_length = False,\\n',\n",
       " '+                 pin_memory = True, no_normalizers = False, device = None):\\n',\n",
       " '+\\n',\n",
       " '+        files = list()\\n',\n",
       " '+        if train:\\n',\n",
       " \"+            partition_files = glob.glob(os.path.join(base_dir, 'train/') + '*/*.mat')\\n\",\n",
       " \"+            #print(f'Adding {len(partition_files)} to dataset.')\\n\",\n",
       " '+            files.extend(partition_files)\\n',\n",
       " '+            \\n',\n",
       " '+        if dev:\\n',\n",
       " \"+            partition_files = glob.glob(os.path.join(base_dir, 'dev/') + '*/*.mat')\\n\",\n",
       " \"+            #print(f'Adding {len(partition_files)} to dataset.')\\n\",\n",
       " '+            files.extend(partition_files)\\n',\n",
       " '+            \\n',\n",
       " '+        if test:\\n',\n",
       " \"+            partition_files = glob.glob(os.path.join(base_dir, 'test/') + '*/*.mat')\\n\",\n",
       " \"+            #print(f'Adding {len(partition_files)} to dataset.')\\n\",\n",
       " '+            files.extend(partition_files)\\n',\n",
       " '+            \\n',\n",
       " '+        self.example_indices = files\\n',\n",
       " '+        self.train = train\\n',\n",
       " '+        self.dev = dev\\n',\n",
       " '+        self.test = test\\n',\n",
       " '+        self.pin_memory = pin_memory \\n',\n",
       " '+        self.device = device\\n',\n",
       " '+\\n',\n",
       " '+        self.example_indices.sort()\\n',\n",
       " '+        np.random.seed(0)\\n',\n",
       " '+        np.random.shuffle(self.example_indices)\\n',\n",
       " '+\\n',\n",
       " '+    \\n',\n",
       " \"+        self.num_speech_features = loadmat(self.example_indices[0])['audio_features'].shape[1]\\n\",\n",
       " \"+        self.num_features        = loadmat(self.example_indices[0])['emg'].shape[1]\\n\",\n",
       " '+        self.limit_length = limit_length\\n',\n",
       " \"+        self.num_sessions = len(glob.glob(os.path.join(base_dir, 'train/') + '*session_*/'))\\n\",\n",
       " '+        \\n',\n",
       " '+        self.no_normalizers = no_normalizers\\n',\n",
       " '+        if not self.no_normalizers:\\n',\n",
       " \"+            self.mfcc_norm, self.emg_norm = pickle.load(open(FLAGS.normalizers_file,'rb'))\\n\",\n",
       " '+\\n',\n",
       " '+    def silent_subset(self):\\n',\n",
       " '+        result = copy(self)\\n',\n",
       " '+        silent_indices = []\\n',\n",
       " '+        for example in self.example_indices:\\n',\n",
       " '+            if example[0].silent:\\n',\n",
       " '+                silent_indices.append(example)\\n',\n",
       " '+        result.example_indices = silent_indices\\n',\n",
       " '+        return result\\n',\n",
       " '+\\n',\n",
       " '+    def subset(self, fraction):\\n',\n",
       " '+        result = copy(self)\\n',\n",
       " '+        result.example_indices = self.example_indices[:int(fraction*len(self.example_indices))]\\n',\n",
       " '+        return result\\n',\n",
       " '+\\n',\n",
       " '+    def __len__(self):\\n',\n",
       " '+        return len(self.example_indices)\\n',\n",
       " '+\\n',\n",
       " '+    @lru_cache(maxsize=None)\\n',\n",
       " '+    def __getitem__(self, i):\\n',\n",
       " '+        \\n',\n",
       " '+        result = loadmat(self.example_indices[i])\\n',\n",
       " '+        \\n',\n",
       " '+        if self.pin_memory:\\n',\n",
       " \"+            keys = ['audio_features',  'emg', 'session_ids', 'raw_emg', 'phonemes', \\n\",\n",
       " \"+                    'parallel_voiced_emg', 'parallel_voiced_audio_features']\\n\",\n",
       " '+            for key in keys:\\n',\n",
       " '+                try:\\n',\n",
       " '+                    result[key] = torch.tensor(result[key].squeeze()).pin_memory()\\n',\n",
       " '+                except:\\n',\n",
       " '+                    continue\\n',\n",
       " '+\\n',\n",
       " '+        return result\\n',\n",
       " '+    \\n',\n",
       " '+\\n',\n",
       " '+    @staticmethod\\n',\n",
       " '+    def collate_raw(batch):\\n',\n",
       " '+        \\n',\n",
       " '+        audio_features        = []\\n',\n",
       " '+        audio_feature_lengths = []\\n',\n",
       " '+        parallel_emg          = []\\n',\n",
       " '+        for ex in batch:\\n',\n",
       " \"+            if ex['silent']:\\n\",\n",
       " \"+                audio_features.append(ex['parallel_voiced_audio_features'])\\n\",\n",
       " \"+                audio_feature_lengths.append(ex['parallel_voiced_audio_features'].shape[0])\\n\",\n",
       " \"+                parallel_emg.append(ex['parallel_voiced_emg'])\\n\",\n",
       " '+            else:\\n',\n",
       " \"+                audio_features.append(ex['audio_features'])\\n\",\n",
       " \"+                audio_feature_lengths.append(ex['audio_features'].shape[0])\\n\",\n",
       " '+                parallel_emg.append(np.zeros(1))\\n',\n",
       " '+                \\n',\n",
       " \"+        phonemes    = [ex['phonemes'] for ex in batch]\\n\",\n",
       " \"+        emg         = [ex['emg'] for ex in batch]\\n\",\n",
       " \"+        raw_emg     = [ex['raw_emg'] for ex in batch]\\n\",\n",
       " \"+        session_ids = [ex['session_ids'] for ex in batch]\\n\",\n",
       " \"+        lengths     = [ex['emg'].shape[0] for ex in batch]\\n\",\n",
       " \"+        silent      = [ex['silent'] for ex in batch]\\n\",\n",
       " '+\\n',\n",
       " \"+        result = {'audio_features':audio_features,\\n\",\n",
       " \"+                  'audio_feature_lengths':audio_feature_lengths,\\n\",\n",
       " \"+                  'emg':emg,\\n\",\n",
       " \"+                  'raw_emg':raw_emg,\\n\",\n",
       " \"+                  'parallel_voiced_emg':parallel_emg,\\n\",\n",
       " \"+                  'phonemes':phonemes,\\n\",\n",
       " \"+                  'session_ids':session_ids,\\n\",\n",
       " \"+                  'lengths':lengths,\\n\",\n",
       " \"+                  'silent':silent}\\n\",\n",
       " '+        \\n',\n",
       " '+        return result \\n',\n",
       " '+    \\n',\n",
       " ' \\n',\n",
       " ' def make_normalizers():\\n',\n",
       " '     dataset = EMGDataset(no_normalizers=True)\\n',\n",
       " 'diff --git a/transduction_model.py b/transduction_model.py\\n',\n",
       " 'index 70b9bf7..7dd6fae 100755\\n',\n",
       " '--- a/transduction_model.py\\n',\n",
       " '+++ b/transduction_model.py\\n',\n",
       " '@@ -5,14 +5,21 @@ import logging\\n',\n",
       " ' import subprocess\\n',\n",
       " ' import random\\n',\n",
       " ' \\n',\n",
       " '+\\n',\n",
       " \"+#libsndfile_path = '/share/software/user/open/libsndfile/1.0.28/lib/libsndfile.so'\\n\",\n",
       " \"+#libsndfile_path = '/share/software/user/open/libsndfile/1.0.28/lib'\\n\",\n",
       " \"+#os.environ['DL_LIBRARY_PATH'] = libsndfile_path + ';' + os.environ['PATH']\\n\",\n",
       " \"+#print(os.environ['PATH'])\\n\",\n",
       " ' import soundfile as sf\\n',\n",
       " ' \\n',\n",
       " '+import scipy\\n',\n",
       " '+\\n',\n",
       " ' import torch\\n',\n",
       " ' from torch import nn\\n',\n",
       " ' import torch.nn.functional as F\\n',\n",
       " ' \\n',\n",
       " '-from read_emg import EMGDataset, SizeAwareSampler\\n',\n",
       " '-from architecture import Model\\n',\n",
       " '+from read_emg import EMGDataset, PreprocessedEMGDataset, SizeAwareSampler, PreprocessedSizeAwareSampler\\n',\n",
       " '+from architecture import Model, S4Model\\n',\n",
       " ' from align import align_from_distances\\n',\n",
       " ' from asr_evaluation import evaluate\\n',\n",
       " ' from data_utils import phoneme_inventory, decollate_tensor, combine_fixed_length\\n',\n",
       " '@@ -20,6 +27,7 @@ from vocoder import Vocoder\\n',\n",
       " ' \\n',\n",
       " ' from absl import flags\\n',\n",
       " ' FLAGS = flags.FLAGS\\n',\n",
       " \"+flags.DEFINE_integer('S4', 0, 'Toggle S4 model in place of transformer')\\n\",\n",
       " \" flags.DEFINE_integer('batch_size', 32, 'training batch size')\\n\",\n",
       " \" flags.DEFINE_integer('epochs', 80, 'number of training epochs')\\n\",\n",
       " \" flags.DEFINE_float('learning_rate', 1e-3, 'learning rate')\\n\",\n",
       " \"@@ -30,6 +38,7 @@ flags.DEFINE_float('data_size_fraction', 1.0, 'fraction of training data to use'\\n\",\n",
       " \" flags.DEFINE_float('phoneme_loss_weight', 0.5, 'weight of auxiliary phoneme prediction loss')\\n\",\n",
       " \" flags.DEFINE_float('l2', 1e-7, 'weight decay')\\n\",\n",
       " \" flags.DEFINE_string('output_directory', 'output', 'output directory')\\n\",\n",
       " \"+flags.DEFINE_string('base_dir', '/oak/stanford/projects/babelfish/magneto/GaddyPaper/processed_data/', 'path to processed EMG dataset')\\n\",\n",
       " ' \\n',\n",
       " ' def test(model, testset, device):\\n',\n",
       " '     model.eval()\\n',\n",
       " '@@ -41,9 +50,9 @@ def test(model, testset, device):\\n',\n",
       " '     seq_len = 200\\n',\n",
       " '     with torch.no_grad():\\n',\n",
       " '         for batch in dataloader:\\n',\n",
       " \"-            X = combine_fixed_length([t.to(device, non_blocking=True) for t in batch['emg']], seq_len)\\n\",\n",
       " \"+            X     = combine_fixed_length([t.to(device, non_blocking=True) for t in batch['emg']], seq_len)\\n\",\n",
       " \"             X_raw = combine_fixed_length([t.to(device, non_blocking=True) for t in batch['raw_emg']], seq_len*8)\\n\",\n",
       " \"-            sess = combine_fixed_length([t.to(device, non_blocking=True) for t in batch['session_ids']], seq_len)\\n\",\n",
       " \"+            sess  = combine_fixed_length([t.to(device, non_blocking=True) for t in batch['session_ids']], seq_len)\\n\",\n",
       " ' \\n',\n",
       " '             pred, phoneme_pred = model(X, X_raw, sess)\\n',\n",
       " ' \\n',\n",
       " '@@ -58,8 +67,8 @@ def test(model, testset, device):\\n',\n",
       " ' def save_output(model, datapoint, filename, device, audio_normalizer, vocoder):\\n',\n",
       " '     model.eval()\\n',\n",
       " '     with torch.no_grad():\\n',\n",
       " \"-        sess = torch.tensor(datapoint['session_ids'], device=device).unsqueeze(0)\\n\",\n",
       " \"-        X = torch.tensor(datapoint['emg'], dtype=torch.float32, device=device).unsqueeze(0)\\n\",\n",
       " \"+        sess  = torch.tensor(datapoint['session_ids'], device=device).unsqueeze(0)\\n\",\n",
       " \"+        X     = torch.tensor(datapoint['emg'], dtype=torch.float32, device=device).unsqueeze(0)\\n\",\n",
       " \"         X_raw = torch.tensor(datapoint['raw_emg'], dtype=torch.float32, device=device).unsqueeze(0)\\n\",\n",
       " ' \\n',\n",
       " '         pred, _ = model(X, X_raw, sess)\\n',\n",
       " '@@ -70,6 +79,8 @@ def save_output(model, datapoint, filename, device, audio_normalizer, vocoder):\\n',\n",
       " '         audio = vocoder(y).cpu().numpy()\\n',\n",
       " ' \\n',\n",
       " '     sf.write(filename, audio, 22050)\\n',\n",
       " '+    #sf.write(filename, audio, 16000)\\n',\n",
       " '+    #scipy.io.wavfile.write(filename, rate22050, audio)\\n',\n",
       " ' \\n',\n",
       " '     model.train()\\n',\n",
       " ' \\n',\n",
       " '@@ -102,13 +113,12 @@ def dtw_loss(predictions, phoneme_predictions, example, phoneme_eval=False, phon\\n',\n",
       " \"     predictions = decollate_tensor(predictions, example['lengths'])\\n\",\n",
       " \"     phoneme_predictions = decollate_tensor(phoneme_predictions, example['lengths'])\\n\",\n",
       " ' \\n',\n",
       " \"-    audio_features = [t.to(device, non_blocking=True) for t in example['audio_features']]\\n\",\n",
       " '-\\n',\n",
       " \"+    audio_features  = [t.to(device, non_blocking=True) for t in example['audio_features']]\\n\",\n",
       " \"     phoneme_targets = example['phonemes']\\n\",\n",
       " ' \\n',\n",
       " '     losses = []\\n',\n",
       " '     correct_phones = 0\\n',\n",
       " '-    total_length = 0\\n',\n",
       " '+    total_length   = 0\\n',\n",
       " \"     for pred, y, pred_phone, y_phone, silent in zip(predictions, audio_features, phoneme_predictions, phoneme_targets, example['silent']):\\n\",\n",
       " '         assert len(pred.size()) == 2 and len(y.size()) == 2\\n',\n",
       " '         y_phone = y_phone.to(device)\\n',\n",
       " '@@ -122,11 +132,9 @@ def dtw_loss(predictions, phoneme_predictions, example, phoneme_eval=False, phon\\n',\n",
       " '             pred_phone = F.log_softmax(pred_phone, -1)\\n',\n",
       " '             phone_lprobs = pred_phone[:,y_phone]\\n',\n",
       " ' \\n',\n",
       " '-            costs = costs + FLAGS.phoneme_loss_weight * -phone_lprobs\\n',\n",
       " '-\\n',\n",
       " '+            costs     = costs + FLAGS.phoneme_loss_weight * -phone_lprobs\\n',\n",
       " '             alignment = align_from_distances(costs.T.cpu().detach().numpy())\\n',\n",
       " '-\\n',\n",
       " '-            loss = costs[alignment,range(len(alignment))].sum()\\n',\n",
       " '+            loss      = costs[alignment,range(len(alignment))].sum()\\n',\n",
       " ' \\n',\n",
       " '             if phoneme_eval:\\n',\n",
       " '                 alignment = align_from_distances(costs.T.cpu().detach().numpy())\\n',\n",
       " '@@ -157,6 +165,7 @@ def dtw_loss(predictions, phoneme_predictions, example, phoneme_eval=False, phon\\n',\n",
       " ' \\n',\n",
       " '     return sum(losses)/total_length, correct_phones/total_length\\n',\n",
       " ' \\n',\n",
       " '+\\n',\n",
       " ' def train_model(trainset, devset, device, save_sound_outputs=True):\\n',\n",
       " '     n_epochs = FLAGS.epochs\\n',\n",
       " ' \\n',\n",
       " '@@ -164,10 +173,22 @@ def train_model(trainset, devset, device, save_sound_outputs=True):\\n',\n",
       " '         training_subset = trainset\\n',\n",
       " '     else:\\n',\n",
       " '         training_subset = trainset.subset(FLAGS.data_size_fraction)\\n',\n",
       " \"-    dataloader = torch.utils.data.DataLoader(training_subset, pin_memory=(device=='cuda'), collate_fn=devset.collate_raw, num_workers=0, batch_sampler=SizeAwareSampler(training_subset, 256000))\\n\",\n",
       " '+        \\n',\n",
       " \"+    #dataloader = torch.utils.data.DataLoader(training_subset, pin_memory=(device=='cuda'), collate_fn=devset.collate_raw, num_workers=0, batch_sampler=SizeAwareSampler(training_subset, 256000))\\n\",\n",
       " '+    \\n',\n",
       " \"+    dataloader = torch.utils.data.DataLoader(training_subset, pin_memory=(device=='cuda'), \\n\",\n",
       " '+                                         collate_fn=devset.collate_raw, num_workers=0, \\n',\n",
       " '+                                         batch_sampler = PreprocessedSizeAwareSampler(training_subset, 256000))\\n',\n",
       " ' \\n',\n",
       " '     n_phones = len(phoneme_inventory)\\n',\n",
       " '-    model = Model(devset.num_features, devset.num_speech_features, n_phones).to(device)\\n',\n",
       " '+    if FLAGS.S4:\\n',\n",
       " '+        model = S4Model(devset.num_features, devset.num_speech_features, n_phones).to(device)\\n',\n",
       " '+    else:\\n',\n",
       " '+        model = Model(devset.num_features, devset.num_speech_features, n_phones).to(device)\\n',\n",
       " '+        \\n',\n",
       " '+    model_parameters = filter(lambda p: p.requires_grad, model.parameters())\\n',\n",
       " '+    params = sum([np.prod(p.size()) for p in model_parameters])\\n',\n",
       " \"+    print('Number of parameters:', params)\\n\",\n",
       " ' \\n',\n",
       " '     if FLAGS.start_training_from is not None:\\n',\n",
       " '         state_dict = torch.load(FLAGS.start_training_from)\\n',\n",
       " '@@ -198,9 +219,9 @@ def train_model(trainset, devset, device, save_sound_outputs=True):\\n',\n",
       " '             optim.zero_grad()\\n',\n",
       " '             schedule_lr(batch_idx)\\n',\n",
       " ' \\n',\n",
       " \"-            X = combine_fixed_length([t.to(device, non_blocking=True) for t in batch['emg']], seq_len)\\n\",\n",
       " \"+            X     = combine_fixed_length([t.to(device, non_blocking=True) for t in batch['emg']], seq_len)\\n\",\n",
       " \"             X_raw = combine_fixed_length([t.to(device, non_blocking=True) for t in batch['raw_emg']], seq_len*8)\\n\",\n",
       " \"-            sess = combine_fixed_length([t.to(device, non_blocking=True) for t in batch['session_ids']], seq_len)\\n\",\n",
       " \"+            sess  = combine_fixed_length([t.to(device, non_blocking=True) for t in batch['session_ids']], seq_len)\\n\",\n",
       " ' \\n',\n",
       " '             pred, phoneme_pred = model(X, X_raw, sess)\\n',\n",
       " ' \\n',\n",
       " '@@ -239,8 +260,12 @@ def main():\\n',\n",
       " ' \\n',\n",
       " '     logging.info(sys.argv)\\n',\n",
       " ' \\n',\n",
       " '-    trainset = EMGDataset(dev=False,test=False)\\n',\n",
       " '-    devset = EMGDataset(dev=True)\\n',\n",
       " '+    #trainset = EMGDataset(dev=False,test=False)\\n',\n",
       " '+    #devset   = EMGDataset(dev=True)\\n',\n",
       " '+    \\n',\n",
       " '+    trainset = PreprocessedEMGDataset(base_dir = FLAGS.base_dir, train = True, dev = False, test = False)\\n',\n",
       " '+    devset   = PreprocessedEMGDataset(base_dir = FLAGS.base_dir, train = False, dev = True, test = False)\\n',\n",
       " '+    \\n',\n",
       " \"     logging.info('output example: %s', devset.example_indices[0])\\n\",\n",
       " \"     logging.info('train / dev split: %d %d',len(trainset),len(devset))\\n\",\n",
       " ' \\n',\n",
       " '\\n',\n",
       " \"['transduction_model.py', '--output_directory=/oak/stanford/projects/babelfish/magneto/GaddyPaper/output/efficiency_sweeps/1', '--epochs', '80', '--data_size_fraction=1', '--hifigan_checkpoint=/oak/stanford/projects/babelfish/magneto/GaddyPaper/pretrained_models/hifigan_finetuned/checkpoint']\\n\",\n",
       " 'output example: /oak/stanford/projects/babelfish/magneto/GaddyPaper/processed_data/dev/session_0/256.mat\\n',\n",
       " 'train / dev split: 8055 200\\n',\n",
       " 'finished epoch 1 - validation loss: 2.2458 training loss: 2.9672 phoneme accuracy: 42.03\\n',\n",
       " 'finished epoch 2 - validation loss: 1.8619 training loss: 2.0238 phoneme accuracy: 53.60\\n',\n",
       " 'finished epoch 3 - validation loss: 1.6366 training loss: 1.7165 phoneme accuracy: 61.24\\n',\n",
       " 'finished epoch 4 - validation loss: 1.5253 training loss: 1.5549 phoneme accuracy: 65.05\\n',\n",
       " 'finished epoch 5 - validation loss: 1.4494 training loss: 1.4515 phoneme accuracy: 67.99\\n',\n",
       " 'finished epoch 6 - validation loss: 1.4506 training loss: 1.3854 phoneme accuracy: 68.82\\n',\n",
       " 'finished epoch 7 - validation loss: 1.3688 training loss: 1.3409 phoneme accuracy: 71.14\\n',\n",
       " 'finished epoch 8 - validation loss: 1.3402 training loss: 1.2970 phoneme accuracy: 72.34\\n',\n",
       " 'finished epoch 9 - validation loss: 1.3793 training loss: 1.2688 phoneme accuracy: 71.77\\n',\n",
       " 'finished epoch 10 - validation loss: 1.3169 training loss: 1.2425 phoneme accuracy: 73.02\\n',\n",
       " 'finished epoch 11 - validation loss: 1.3068 training loss: 1.2138 phoneme accuracy: 74.29\\n',\n",
       " 'finished epoch 12 - validation loss: 1.2572 training loss: 1.1944 phoneme accuracy: 75.64\\n',\n",
       " 'finished epoch 13 - validation loss: 1.3165 training loss: 1.1801 phoneme accuracy: 74.34\\n',\n",
       " 'finished epoch 14 - validation loss: 1.2584 training loss: 1.1604 phoneme accuracy: 75.77\\n',\n",
       " 'finished epoch 15 - validation loss: 1.2355 training loss: 1.1458 phoneme accuracy: 76.55\\n',\n",
       " 'finished epoch 16 - validation loss: 1.2363 training loss: 1.1381 phoneme accuracy: 76.73\\n',\n",
       " 'finished epoch 17 - validation loss: 1.2596 training loss: 1.1235 phoneme accuracy: 76.54\\n',\n",
       " 'finished epoch 18 - validation loss: 1.2403 training loss: 1.1091 phoneme accuracy: 76.98\\n',\n",
       " 'finished epoch 19 - validation loss: 1.2534 training loss: 1.0982 phoneme accuracy: 77.14\\n',\n",
       " 'finished epoch 20 - validation loss: 1.1975 training loss: 1.0949 phoneme accuracy: 78.36\\n',\n",
       " 'finished epoch 21 - validation loss: 1.2287 training loss: 1.0838 phoneme accuracy: 78.90\\n',\n",
       " 'finished epoch 22 - validation loss: 1.1856 training loss: 1.0720 phoneme accuracy: 79.28\\n',\n",
       " 'finished epoch 23 - validation loss: 1.2160 training loss: 1.0633 phoneme accuracy: 77.64\\n',\n",
       " 'finished epoch 24 - validation loss: 1.1738 training loss: 1.0578 phoneme accuracy: 79.82\\n',\n",
       " 'finished epoch 25 - validation loss: 1.1974 training loss: 1.0547 phoneme accuracy: 79.14\\n',\n",
       " 'finished epoch 26 - validation loss: 1.1923 training loss: 1.0471 phoneme accuracy: 80.24\\n',\n",
       " 'finished epoch 27 - validation loss: 1.1754 training loss: 1.0396 phoneme accuracy: 79.87\\n',\n",
       " 'finished epoch 28 - validation loss: 1.1642 training loss: 1.0260 phoneme accuracy: 80.12\\n',\n",
       " 'finished epoch 29 - validation loss: 1.1703 training loss: 1.0252 phoneme accuracy: 80.19\\n',\n",
       " 'finished epoch 30 - validation loss: 1.1701 training loss: 1.0163 phoneme accuracy: 79.68\\n',\n",
       " 'finished epoch 31 - validation loss: 1.1747 training loss: 1.0127 phoneme accuracy: 80.32\\n',\n",
       " 'finished epoch 32 - validation loss: 1.1540 training loss: 1.0055 phoneme accuracy: 80.84\\n',\n",
       " 'finished epoch 33 - validation loss: 1.1263 training loss: 1.0014 phoneme accuracy: 81.47\\n',\n",
       " 'finished epoch 34 - validation loss: 1.1609 training loss: 0.9978 phoneme accuracy: 80.86\\n',\n",
       " 'finished epoch 35 - validation loss: 1.1524 training loss: 0.9921 phoneme accuracy: 81.37\\n',\n",
       " 'finished epoch 36 - validation loss: 1.1443 training loss: 0.9835 phoneme accuracy: 81.59\\n',\n",
       " 'finished epoch 37 - validation loss: 1.1330 training loss: 0.9833 phoneme accuracy: 81.48\\n',\n",
       " 'finished epoch 38 - validation loss: 1.1504 training loss: 0.9749 phoneme accuracy: 81.94\\n',\n",
       " 'finished epoch 39 - validation loss: 1.1353 training loss: 0.9676 phoneme accuracy: 82.06\\n',\n",
       " 'finished epoch 40 - validation loss: 1.0707 training loss: 0.9056 phoneme accuracy: 84.27\\n',\n",
       " 'finished epoch 41 - validation loss: 1.0597 training loss: 0.8852 phoneme accuracy: 84.71\\n',\n",
       " 'finished epoch 42 - validation loss: 1.0650 training loss: 0.8749 phoneme accuracy: 84.23\\n',\n",
       " 'finished epoch 43 - validation loss: 1.0697 training loss: 0.8685 phoneme accuracy: 84.54\\n',\n",
       " 'finished epoch 44 - validation loss: 1.0759 training loss: 0.8648 phoneme accuracy: 84.42\\n',\n",
       " 'finished epoch 45 - validation loss: 1.0736 training loss: 0.8631 phoneme accuracy: 84.61\\n',\n",
       " 'finished epoch 46 - validation loss: 1.0695 training loss: 0.8599 phoneme accuracy: 84.51\\n',\n",
       " 'finished epoch 47 - validation loss: 1.0832 training loss: 0.8536 phoneme accuracy: 84.48\\n',\n",
       " 'finished epoch 48 - validation loss: 1.0608 training loss: 0.8263 phoneme accuracy: 85.61\\n',\n",
       " 'finished epoch 49 - validation loss: 1.0631 training loss: 0.8168 phoneme accuracy: 85.39\\n',\n",
       " 'finished epoch 50 - validation loss: 1.0559 training loss: 0.8091 phoneme accuracy: 85.77\\n',\n",
       " 'finished epoch 51 - validation loss: 1.0707 training loss: 0.8075 phoneme accuracy: 85.61\\n',\n",
       " 'finished epoch 52 - validation loss: 1.0690 training loss: 0.8037 phoneme accuracy: 85.77\\n',\n",
       " 'finished epoch 53 - validation loss: 1.0583 training loss: 0.7997 phoneme accuracy: 86.02\\n',\n",
       " 'finished epoch 54 - validation loss: 1.0688 training loss: 0.7998 phoneme accuracy: 85.93\\n',\n",
       " 'finished epoch 55 - validation loss: 1.0640 training loss: 0.7959 phoneme accuracy: 85.87\\n',\n",
       " 'finished epoch 56 - validation loss: 1.0747 training loss: 0.7917 phoneme accuracy: 85.66\\n',\n",
       " 'finished epoch 57 - validation loss: 1.0616 training loss: 0.7771 phoneme accuracy: 85.95\\n',\n",
       " 'finished epoch 58 - validation loss: 1.0648 training loss: 0.7716 phoneme accuracy: 86.08\\n',\n",
       " 'finished epoch 59 - validation loss: 1.0648 training loss: 0.7694 phoneme accuracy: 85.83\\n',\n",
       " 'finished epoch 60 - validation loss: 1.0710 training loss: 0.7660 phoneme accuracy: 86.14\\n',\n",
       " 'finished epoch 61 - validation loss: 1.0707 training loss: 0.7646 phoneme accuracy: 86.27\\n',\n",
       " 'finished epoch 62 - validation loss: 1.0743 training loss: 0.7635 phoneme accuracy: 85.99\\n',\n",
       " 'finished epoch 63 - validation loss: 1.0711 training loss: 0.7563 phoneme accuracy: 86.36\\n',\n",
       " 'finished epoch 64 - validation loss: 1.0739 training loss: 0.7532 phoneme accuracy: 86.19\\n',\n",
       " 'finished epoch 65 - validation loss: 1.0732 training loss: 0.7515 phoneme accuracy: 86.43\\n',\n",
       " 'finished epoch 66 - validation loss: 1.0750 training loss: 0.7510 phoneme accuracy: 86.27\\n',\n",
       " 'finished epoch 67 - validation loss: 1.0674 training loss: 0.7486 phoneme accuracy: 86.38\\n',\n",
       " 'finished epoch 68 - validation loss: 1.0820 training loss: 0.7474 phoneme accuracy: 86.45\\n',\n",
       " 'finished epoch 69 - validation loss: 1.0783 training loss: 0.7446 phoneme accuracy: 86.56\\n',\n",
       " 'finished epoch 70 - validation loss: 1.0785 training loss: 0.7428 phoneme accuracy: 86.48\\n',\n",
       " 'finished epoch 71 - validation loss: 1.0805 training loss: 0.7419 phoneme accuracy: 86.47\\n',\n",
       " 'finished epoch 72 - validation loss: 1.0806 training loss: 0.7408 phoneme accuracy: 86.52\\n',\n",
       " 'finished epoch 73 - validation loss: 1.0775 training loss: 0.7403 phoneme accuracy: 86.59\\n',\n",
       " 'finished epoch 74 - validation loss: 1.0815 training loss: 0.7402 phoneme accuracy: 86.56\\n',\n",
       " 'finished epoch 75 - validation loss: 1.0776 training loss: 0.7387 phoneme accuracy: 86.59\\n',\n",
       " 'finished epoch 76 - validation loss: 1.0768 training loss: 0.7364 phoneme accuracy: 86.64\\n',\n",
       " 'finished epoch 77 - validation loss: 1.0785 training loss: 0.7372 phoneme accuracy: 86.62\\n',\n",
       " 'finished epoch 78 - validation loss: 1.0798 training loss: 0.7368 phoneme accuracy: 86.58\\n',\n",
       " 'finished epoch 79 - validation loss: 1.0797 training loss: 0.7371 phoneme accuracy: 86.56\\n',\n",
       " 'finished epoch 80 - validation loss: 1.0811 training loss: 0.7355 phoneme accuracy: 86.57\\n',\n",
       " \"targets: ['they were described as vast spiderlike machines nearly a hundred feet high capable of the speed of an express train and able to shoot out a beam of intense heat masked batteries chiefly of field guns had been planted in the country about horsell common and especially between the woking district and london', 'at the sight of the second monster my nerve went altogether', 'when i returned to the common the sun was setting', 'the people who landed there from the boats went tramping off down the lane', 'the sudden near concussion the last close upon the first made my heart jump', 'i told him it was the heliograph signallingthat it was the sign of human help and effort in the sky', 'so heavy was the stress of the storm just at this place that i had the hardest task to win my way up the hill', 'the martians appear to be moving slowly towards chertsey or windsor', 'it was the strangest spectacle that black expanse set with fire', 'it was seen early in the morning rushing over winchester eastward a line of flame high in the atmosphere', 'the ordinary sapper is a great deal better educated than the common soldier and they discussed the peculiar conditions of the possible fight with some acuteness', 'what i saw of the destruction of weybridge and shepperton', 'the storm had left the sky clear and over the smoke of the burning land the little fading pinpoint of mars was dropping into the west when a soldier came into my garden', 'in spite of all that has happened since i still remember that vigil very distinctly the black and silent observatory the shadowed lantern throwing a feeble glow upon the floor in the corner the steady ticking of the clockwork of the telescope the little slit in the roofan oblong profundity with the stardust streaked across it', 'the afternoon papers puffed scraps of news under big headlines', 'are we far from sunbury i said in a matteroffact tone', 'i was immediately the self of every day againa decent ordinary citizen', 'it was heavy this vapour heavier than the densest smoke so that after the first tumultuous uprush and outflow of its impact it sank down through the air and poured over the ground in a manner rather liquid than gaseous abandoning the hills and streaming into the valleys and ditches and watercourses even as i have heard the carbonicacid gas that pours from volcanic clefts is wont to do', 'book one the coming of the martians', 'i saw and thought nothing of the other four martian monsters my attention was riveted upon the nearer incident', 'he gave a cry of terror sprang sideways and rushed on before i could gather my wits sufficiently to speak to him', 'the exodus from london', 'like burnt meat', 'there were lights in the upper windows of the houses as the people went to bed', 'after breakfast instead of working i decided to walk down towards the common', 'quickly one after the other one two three four of the armoured martians appeared far away over the little trees across the flat meadows that stretched towards chertsey and striding hurriedly towards the river', 'the living intelligence the martian within the hood was slain and splashed to the four winds of heaven and the thing was now but a mere intricate device of metal whirling to destruction', 'they will be hot under foot for days on account of the thick soil of pine needles and turf he said and then grew serious over poor ogilvy', 'one has to imagine as well as one may the fate of those batteries towards esher waiting so tensely in the twilight', 'a man in a blue and white blazer addressed my brother full of strange tidings', 'unable from his window to learn what was happening my brother went down and out into the street just as the sky between the parapets of the houses grew pink with the early dawn', 'xi', 'im agoin ome i am', 'some way behind these came a man in workday clothes riding one of those oldfashioned tricycles with a small front wheel', 'i expected nothing but death', 'the soldiers had made the people on the outskirts of horsell lock up and leave their houses', 'we have chanced to come in for the thick of it said i and that is all', 'keep back said several', 'he told me that a faint stirring was occasionally still audible within the case but that the workmen had failed to unscrew the top as it afforded no grip to them', 'i rose and walked unsteadily up the steep incline of the bridge', 'then suddenly he noticed with a start that some of the grey clinker the ashy incrustation that covered the meteorite was falling off the circular edge of the end', 'he walked from westminster to his apartments near regents park about two', 'my dear wifes sweet anxious face peering at me from under the pink lamp shade the white cloth with its silver and glass table furniturefor in those days even philosophical writers had many little luxuriesthe crimsonpurple wine in my glass are photographically distinct', 'i saw that the driving clouds had been pierced as it were by a thread of green fire suddenly lighting their confusion and falling into the field to my left', 'but very early in the morning poor ogilvy who had seen the shooting star and who was persuaded that a meteorite lay somewhere on the common between horsell ottershaw and woking rose early with the idea of finding it', 'in london also on saturday night there was a thunderstorm and my brother reached waterloo in a cab', 'that last stage of exhaustion which to us is still incredibly remote has become a presentday problem for the inhabitants of mars', 'later a second company marched through chobham to deploy on the north side of the common', 'they may keep the pit and kill people who come near them but they cannot get out of it', 'henderson was a minute or so taking it in', 'i talked with these soldiers for a time i told them of my sight of the martians on the previous evening', 'the decapitated colossus reeled like a drunken giant but it did not fall over', 'once i had turned i did not dare to look back', 'my imagination was full of those striding metallic monsters and of the dead body smashed against the fence', 'for a long time i drifted so painful and weary was i after the violence i had been through and so intense the heat upon the water', 'and presently men selling unnaturally early newspapers came bawling into the street', 'the hussars had been on the road beyond the curvature of the ground and he saw nothing of them', 'the martians are able to discharge enormous clouds of a black and poisonous vapour by means of rockets', 'had they prepared pitfalls', 'the martians took as much notice of such advances as we should of the lowing of a cow', 'such an extraordinary effect in unmanning me it had that i ran weeping silently as a child might do', 'a man too frightened to drop the portmanteau he carried on his shoulder swung round and sent me staggering with a blow from the corner of his burden', 'the public would be fairly warned of the approach of danger and elaborate measures were being taken for the protection of the people in the threatened southwestern suburbs', 'were they intelligent mechanisms', 'it was too far for me to recognise anyone there but afterwards i learned that ogilvy stent and henderson were with others in this attempt at communication', 'the martians had but to advance into the line of fire and instantly those motionless black forms of men those guns glittering so darkly in the early night would explode into a thunderous fury of battle', 'he was turned aside by the fire and hid among some almost scorching heaps of broken wall as one of the martian giants returned', 'in another moment a bend in the road hid the three men from my brothers eyes', 'as it bulged up and caught the light it glistened like wet leather', 'the smoke already extended far away to the east and westto the byfleet pine woods eastward and to woking on the west', 'vertical black figures in twos and threes would advance stop watch and advance again spreading out as they did so in a thin irregular crescent that promised to enclose the pit in its attenuated horns', 'what news', 'then came a crying from the common', 'my wife at least did not find my experience incredible', 'the top had certainly ceased to rotate', 'keep back', 'theres fighting going on about weybridge was the extent of their information', 'he was now very anxious on my account and disturbed at the evident magnitude of the trouble', 'i startled my wife at the doorway so haggard was i i went into the dining room sat down drank some wine and so soon as i could collect myself sufficiently i told her the things i had seen', 'the heatray had shaved the chimney tops and passed', 'after a fruitless struggle to get aboard a northwestern train at chalk farmthe engines of the trains that had loaded in the goods yard there ploughed through shrieking people and a dozen stalwart men fought to keep the crowd from crushing the driver against his furnacemy brother emerged upon the chalk farm road dodged across through a hurrying swarm of vehicles and had the luck to be foremost in the sack of a cycle shop', 'i went in again according to my promise to get my servants box lugged it out clapped it beside her on the tail of the dog cart and then caught the reins and jumped up into the drivers seat beside my wife', 'i was still standing on the bank by the side of the road', 'the crowd about the pit had increased and stood out black against the lemon yellow of the skya couple of hundred people perhaps', 'it is curious to recall some of the mental habits of those departed days', 'for a time he sat in blank silence', 'dense clouds of smoke or dust visible through a powerful telescope on earth as little grey fluctuating patches spread through the clearness of the planets atmosphere and obscured its more familiar features', 'i explained hastily that i had to leave my home and so secured the dog cart', 'youd better go along and report yourself to brigadiergeneral marvin and tell him all you know', 'the only offensive operation men ventured upon after that night was the preparation of mines and pitfalls and even in that their energies were frantic and spasmodic', 'over that above the blackened heather and ruined buildings that stretched far and wide stood one as sentinel while the rest abandoned their vast fightingmachines and descended into the pit', '12 of the horse artillery', 'of course the two were quite unable to do anything', 'well said the lieutenant i suppose its my business to see it too', 'nothing was to be seen save flat meadows cows feeding unconcernedly for the most part and silvery pollard willows motionless in the warm sunlight', 'i was explainin these is vallyble', 'i described the heatray to them and they began to argue among themselves', 'it was as if each man were suddenly and momentarily turned to fire', 'i fell and lay still', 'he was a driver in the artillery and had only come into action about seven', 'what confounded nonsense', 'when my brother looked again they were retreating', 'are we far from sunbury', 'i heard the crackle of fire in the sandpits and the sudden squeal of a horse that was as suddenly stilled', 'its amovin he said to me as he passed ascrewin and ascrewin out', 'the fear i felt was no rational fear but a panic terror not only of the martians but of the dusk and stillness all about me', 'yet across the gulf of space minds that are to our minds as ours are to those of the beasts that perish intellects vast and cool and unsympathetic regarded this earth with envious eyes and slowly and surely drew their plans against us', 'ripley street was deserted and except for a lighted window or so the village showed not a sign of life but i narrowly escaped an accident at the corner of the road to pyrford where a knot of people stood with their backs to me', 'never once after the martian at st georges hill was brought down did they give the artillery the ghost of a chance against them', 'its a cylinderan artificial cylinder man', 'they both stopped in sight down the lane where the third man lay insensible', 'the less courageous of the robbers made off and his companion followed him cursing his cowardice', 'but looking i presently saw something stirring within the shadow greyish billowy movements one above another and then two luminous diskslike eyes', 'coming out he bought a referee', 'he managed to get alive into the ditch by the side of the road and so escaped to woking', 'he noticed an unusual number of police regulating the traffic', 'the noise was deafening', 'i ran for i perceived that in a moment everyone upon this side of the hill would be moving', 'it was strange to see the place quite tranquil quite desolate under the hot blue sky with the smoke and little threads of flame going straight up into the heat of the afternoon', 'one or two trains came in from richmond putney and kingston containing people who had gone out for a days boating and found the locks closed and a feeling of panic in the air', 'this jet of fire had become invisible about a quarter past twelve', 'a large portion of the cylinder had been uncovered though its lower end was still embedded', 'hidden by a pine wood as they were they seem to have been quite unsuspected by the martian nearest to them', 'hats and dresses caught fire', 'during the opposition of 1894 a great light was seen on the illuminated part of the disk first at the lick observatory then by perrotin of nice and then by other observers', 'killed he said staring about him', 'here and there was a burning bush or tree', 'as he talked things about us came darkly out of the darkness and the trampled bushes and broken rose trees outside the window grew distinct', 'they had got hold of a little hand truck and were piling it up with uncleanlooking bundles and shabby furniture', 'that flicker in the sky tells of the gathering storm', 'good lord said henderson', 'i saw a young man a shop assistant in woking i believe he was standing on the cylinder and trying to scramble out of the hole again', 'the milkman came as usual', 'whos there he said also whispering standing under the window and peering up', 'not stopping to look again i wrenched the horses head hard round to the right and in another moment the dog cart had heeled over upon the horse the shafts smashed noisily and i was flung sideways and fell heavily into a shallow pool of water', 'my god he said as i drew him in', 'i saw a lad trundling off the barrow of apples', 'they were all too assiduously engaged to talk to us as we passed', 'the majority of people in london do not read sunday papers', 'for a moment perhaps i could not clearly understand how i came there', 'where the road grows narrow and black between the high banks the crowd jammed and a desperate struggle occurred', 'i looked again at the martian and saw he was now moving eastward along the riverbank with a swift rolling motion', 'were the powder mills at hounslow ready as a snare', 'he approached the mass surprised at the size and more so at the shape since most meteorites are rounded more or less completely', 'only this morning i officiated at early celebration', 'this was the second cylinder', 'it seems there were a few people alive there frantic for the most part and many burned and scalded', 'after that no body of men would stand against them so hopeless was the enterprise', 'henderson and ogilvy were not there', 'one of our chimneys cracked as if a shot had hit it flew and a piece of it came clattering down the tiles and made a heap of broken red fragments upon the flower bed by my study window', 'not far from me was a little oneroomed squatters hut of wood surrounded by a patch of potato garden', 'then it was and then only that he realised something of the full power and terror of these monsters', 'save that an unknown element giving a group of four lines in the blue of the spectrum is concerned we are still entirely ignorant of the nature of this substance', 'at that i gripped my wifes arm and without ceremony ran her out into the road', 'the occasional howling of the martians had ceased they took up their positions in the huge crescent about their cylinders in absolute silence', 'what hasnt in the obscurity i could see he made a gesture of despair', 'there was no sensible transition from one state of mind to the other', 'as i did so a second report followed and a big projectile hurtled overhead towards hounslow', 'the night was warm and still and a little oppressive the sound of guns continued intermittently and after midnight there seemed to be sheet lightning in the south', 'in london that night poor hendersons telegram describing the gradual unscrewing of the shot was judged to be a canard and his evening paper after wiring for authentication from him and receiving no replythe man was killeddecided not to print a special edition', 'i got up almost mechanically went into the dining room and drank some whisky and then i was moved to change my clothes', 'london in danger of suffocation', 'they became pillars of bloodshot smoke at the first touch of day', 'talk about fishers of menfighters of fish it is this time', 'the respectable inhabitants of the place men in golf and boating costumes wives prettily dressed were packing riverside loafers energetically helping children excited and for the most part highly delighted at this astonishing variation of their sunday experiences', 'i do not clearly remember the arrival of the curate so that probably i dozed', 'there is one thing i said to allay the fears i had aroused they are the most sluggish things i ever saw crawl', 'we got to leatherhead without misadventure about nine oclock and the horse had an hours rest while i took supper with my cousins and commended my wife to their care', 'would that i had', 'he returned and tried in vain to divert his attention to his examination notes', 'there was some booing from those more thoughtless and excitable souls to whom a crowd is always an occasion for noise and horseplay', 'with infinite complacency men went to and fro over this globe about their little affairs serene in their assurance of their empire over matter', 'he was immensely excited at the news and in the excess of his feelings invited me up to take a turn with him that night in a scrutiny of the red planet', 'in the centre sticking into the skin of our old planet earth like a poisoned dart was this cylinder', 'apparently his neck had been broken', 'halfway through the lieutenant interrupted him and looked up at me', 'air was either entering or escaping at the rim with a thin sizzling sound', 'it had occurred towards midnight of the twelfth and the spectroscope to which he had at once resorted indicated a mass of flaming gas chiefly hydrogen moving with an enormous velocity towards this earth', 'so ive locked up my house and come on', 'it seemed to me that the pit had been enlarged and ever and again puffs of vivid green vapour streamed up and out of it towards the brightening dawnstreamed up whirled broke and vanished', 'but so little it was so silvery warma pins head of light', 'i screamed aloud and scalded half blinded agonised i staggered through the leaping hissing water towards the shore', 'i struggled to my feet at last and crouching and making use of every chance of cover i made a run for this', 'we went down the lane by the body of the man in black sodden now from the overnight hail and broke into the woods at the foot of the hill', 'this they did with the vapour banks near us as we saw in the starlight from the window of a deserted house at upper halliford whither we had returned', 'why the shots ceased after the tenth no one on earth has attempted to explain', 'there was a noise of business from the gasworks and the electric lamps were all alight', 'i began walking therefore in a big curve seeking some point of vantage and continually looking at the sandheaps that hid these newcomers to our earth', 'he was deaf in one ear', 'it was the head of the shopman who had fallen in but showing as a little black object against the hot western sun', 'by a common impulse we clambered higher', 'they stopped at the square as if undecided which way to take and finally turned eastward along the strand', 'the nature of the accident he could not ascertain indeed the railway authorities did not clearly know at that time', 'in addition ogilvys wire to the astronomical exchange had roused every observatory in the three kingdoms', 'the thought of the confined creature was so dreadful to him that he forgot the heat and went forward to the cylinder to help turn', 'what dye meana gun', 'in the rest of the world the stream of life still flowed as it had flowed for immemorial years', 'he was passed by a number of cyclists some horsemen and two motor cars', 'i talked to her reassuringly pointing out that the martians were tied to the pit by sheer heaviness and at the utmost could but crawl a little out of it but she answered only in monosyllables']\\n\",\n",
       " \"predictions: [' there was scribed as vast spiraling of machines nearly 100 feet high capable of the speed of express treating and able to stoop out a maim of a tinsed heat masked batteries chively of field guns had been planted in the country about horse of common and has been to leave between the walking district and london', ' at the sight of the second monster my turf went altogether', ' who i returned to the comment when the sun was setting', ' the people who landed there from the boats went driving off down the lane', ' the sudden near the kitchen the last close upon the first made my heart shun', ' i told him it was the photograph signaling that it was the sign of him held in effort of this guy', ' so if it was the stress of the storm just at this place that i hand the artists desk to wind my way up the hill', ' the marshes may be moving slowly towards shirt c or wind sir', ' it was strange spectacle that black and expanse sent with fire', ' and once seen early in the morning rushing for winchester eastward a line of flame high in the atmosphere', ' the ordinary sapper has a great deal of matter educated from the common soldier and they discussed the peculiar conditions of the possible fight with some acuteness', ' what i saw of the destruction of weybridge and shepperton', ' the storm had left this ike layer and over the smoke of the burning land the little fading minpoint of mars was dropping into the west when a soldier came into my regard', ' in spite of all that has happened since i still remember that vigil very instinctively the black and silent observatory the shadowed lantern throwing a feeble clove at the floor in the corner the semitine of the clockwork of the telescom the little sleet in the rove and an upbump brofundity with the stardust streaked across it', ' the afterearth papers bump the scraps of twos under big inlines', ' ill be far from sunbury i said in a batterafactodurn', ' i was immediately the self of memory a again and he stood ordinary citizen', ' it was heavy this paper after that the de ensis spoke so that after the first tumultuous uprush and outflow of its embact he sank down through the air and bore over the ground in a matter rather liquid that cashes amanding the hills and streaming into the valleys and ditches and watercourses even as everd the carpal cay acid gas that pours from volcanic cliffs is wont to do the', ' dokuan the coming of the martians', ' i saw and thought nothing of the other four martian monsters by intention was riveted upon the deirior incident', ' he gave a cry of terror spraying sideways and rushed on before i could gather my wits sufficiently to speak to him', ' the exodus from london', ' like part meat', ' there were lights in the upper windows of the houses as the people went to men', ' after breakfast instead of working and a sudden to walk down towards the common', ' quickly one after the other one two three four of the armored martians appeared far away over the little trees across the flat meadows that stretched towards chertsey and striding hurriedly towards the river', ' the living intelligence the martian within the hood was the lane and splashed to the four winds of eden and the thing was now but a mere integrated vice of betel whirling due to destruction', ' that will be hot underfoot for days on account of the thing soil of eindales endurve he said and then grew serious over four ogilvii', ' one has to imagine as well as one may to vent those batteries towards esther we are waiting so to see in the twilight', ' a man in a blue and white placer addressed my brother in a full of strange lettings', ' unable from his window to alert what was happening my brother went down and down into the street he just has this tie between the puffins of the houses group bank with the early tom', ' 11', ' id be going home by him', ' some lay behind these king of man and worked enclose riding one of those oldfashioned tricycles with a small front wheel', ' i expected nothing but death', ' the soldiers and maid the people on the outskirts of horssell lock up and leave their houses', ' we have a chance to come and for the sake of it said i and that is all', ' keep back sansa froh', ' he told me that a faint starting was occasionally still on a pole with the case but that the workmen and face it unscrew the top as it afforded to grip to them', ' i rose and walked unstudily up this deep and conned of the bridge', ' then suddenly aortes with a start that some of the grey clinker to the azhi and crustacean that covered the meteorite was falling off the particular hedge of the endon', ' he walked from westminster to his apartments and he reached his park about two', ' my here wights sway into his face veering at me from much the big lampretained white day of love within sober and classteam of fleerture for in those days even a fellow typical writers many little luxuries the cripsum people wide by in my class are foodgravically distinct', ' i saw that the driving clouds had been pierced as it were by a threat of green fire suddenly lighting their confusion and falling into the field to my left', ' but very early in the morning poor ogilvie who had seen the chinese star and who was persuaded that a meteor it lay somewhere on the common between a poor cell on her shelf invoking rose certainly with the idea of finding it', ' in london also on saturday night there was a little storm and my brother reached waterloo in a cab', ' that last stand of exhaustion which was is still in chronopyrabote and become present in problem for the evidence of mars', ' later a second company bortz dshavam flew on the north side of the common', ' they men keep them in and kill people who can leave them but they cannot get out of it', ' anderson was a minute or so hanging', ' i talked with these soldiers for a time and told them of my sight of the martians on the previous evening', ' the decapitating colossus reeled like a drunken giant but it did not fall over', ' it was my intern i did not dare and i looked back', ' my imagination was full of those striding in metallica monsters and of the dead body smashed against the vents', ' for a long time a drip in so painfully and weary was i after the violence i had been through and so indent these upon the water', ' and presently mays selling of undoctrally early newspapers came falling into the street', ' the hustlers had been on the road via the curvature of the ground and he saw nothing of them', ' the martians are able to discharge enormous clouds of a black and poisonous vapor by base of rockets', ' and they prepared it in vintfalls', ' the martian took as much dose of flanctobvastas as we should of the loaning of a cow', ' such extraordinary effect and unpanding me it had that i ran weeping silently as a child might too', ' a man to fight tropled the portament he carried on his shoulder swam round and sent a standring with a blow from the corner of his burden', ' a public would be fairly warned of the frigid danger and elaborate measures were being taken for the brexit of the people and the threatened southwestern stophos', ' were the intelligent mechanisms', ' it was no far from me to recognize anyone there but afterwards i learned that ogillley stent and henderson were with others in this attempt and to mutation', ' martians had not to advance into the light of fire and instantly the smoosh this black forms of men those guns clearing so darkly in the early night one explode into a thunders fury of battle', ' he was turned aside by the fire and it among some almost scorching heaps broken wall as one of the martians giants returned', ' in a little bit of it in the road in the three men from my brothers size', ' as it bulged up and caught the light into glistened languette valour', ' the smoke while running sent it far away to the east and west to the bifleet pine woods eastward into walking on the quest', ' frugal black figures and twos and threes would advance stop watch and advance again spreading out as they did so in the then irregular crescent that promised to envelop the pit in its untentuated oars', ' nasa news', ' then keep a crying from the con', ' my wife at east did not find my experience incredible', ' that happened certainly ceased to corrode', ' keep back', ' this fighting going on about weybridge was the extent of their formation', ' he was so very anxious on my account and disturbed at the evident lactatune of the creature', ' i started with my wife at the doorway so a hacker was i i went to the dining room sat down drained the bride and so soon as i could collect myself eventually i told her the things i had seen', ' the heat ray achieved should be tough at best', ' after a fruitless struggle to cant afford a northwest refrain and shock farm the distance of the trains that timte loaded in the good guards there blowed through a shrinking people and a dozen soursmen found steam in the crowd from crashing the tribor against this furnished my brother and her upon the shock farmer road he marched across through a hurrying storm of vehicles and down the luck of me the foremost in the sack of a cycle shop', ' i went in again according to my promise hed get my service box looked and out clapped beside her on the tail of the dog cart and had gone to the reins and shoved up into the drivers seat beside my wife', ' im still standing on the bank by the side of the road', ' the crowd above the pinté inquiries and stood out black against the lemon dianote of the sky a couple of hundred people perhaps', ' in a security strangle are some of the mental habits of those infernive days', ' for a time he set in blank silence', ' its clouds of smoke or dust visible through a powerful telescope on earth as little gray fluctuating madness spread through the clearness of the planets atmosphere and obscured as more familiar features', ' i explained hastily that i had to leave my house and so secured the dock cart', ' youd better go along and report shev to prepare gerro burban and tell him all you know', ' the only of it in preparation an infatured pawn after that night with the preparation of myse and fitfalls and even in that their engine knees were franding in spasmonic', ' over that above the blackened gather and ruined buildings that stretched afar at y so what as sintl where the rest abandoned the fastfiring machines and descended into the pit', ' 12 of the horse artillery', ' of course the two were quite unable to do anything', ' well still it did as a boat as my business to see and knew', ' nothing was of misting save flat meadows cows of feeding unconcertedly for the most part and silky molar wallows marshalless in the warm sunlight', ' i was explaining these its valuable', ' i described the heatrayed them and they began to octor among themselves', ' it was as if each man were suddenly and momentarily turned to fire', ' i fell at leiston', ' i was a driver in the artillery and had only come in tactional 7', ' i want the confounded guns and', ' when my brother looked again they were retreating', ' ill be fired from st barry', ' i heard the crackle of fire in the sand pits and the sudden squeal of a horse that was stashed suddenly still', ' its moving he said to me as he passed a screwing and a screwing dowd', ' the fear i felt was no rational fear but a panic dare not only of the martians but of the dusk and stillness all about me', ' and yet across the coldless space mind and error to our minds as ours are to those of the beast that perish into lengths vast and cool and unsympathetic are incarnate this earth within the size and slowly and surely true that plants against us', ' ripley street was deserted and except for a light it went or so the village showed not a sideflife but i terribly escaped an accident at the corner of the road to pyreford where not a people stood with their backs to me', ' never once after the martian at st georges hill was won down did they give the artillery the course of a chance against them', ' its a cylinder an artificial cylinder man', ' they both stopped in sight down the lane with a third lay insensible', ' the last courageous of the ruffus of madoff and his companion to followed him cursing has cowardice', ' mutt of looking abresively saw something stirring within the shadow graysch veloey movements one of another and the thin two openness and thin six dark eyes', ' you can count he but a referee', ' he managed to look into the ditch by the sun and so escaped to walking', ' he doesed an unusual number of police regulating the trapping', ' the noise was deafening', ' i ran from a seat that in a moment everyone upon the side of the hill would be moving', ' it was strange to see the place break to tranquil quite testily to an earthhot blue sky with the spook into null the rest of flame going straight up into the heat of the everdew', ' what a duels trains came in from richmond bonday and kingston containing people who had gone out for a days morning and found the locks close and a feeling of panic in there', ' the jet of fire had become invisible upon a cornermessed clove', ' a large portion of the cylinder had been unencoupered though a slower end was still unmeted', ' by a pine wood as they were they seemed to have been quite as nonsuspected by the martian nearest to them', ' and its indresses go on fire', ' during the opposition of ad dunny iv a great light was seen on the illuminated part of the daisque first at the lake observatory then at the ayrton of nies and then by other observers', ' he killed he said staring about him', ' here and there was a parting bush or tree', ' as he talked thanks about us came thoroughly out of the darkness and the traveled footages and broken road trees outside the window grew distinct', ' they had got hold of a little hand truck and were filing a handtop with uncleanlicking models and shabby furniture', ' then to flicker in the sky the tales of the gathering storm', ' a good lord st anderson', ' i saw a beaten man a shop assisted in walking a belitti bus standing over the cylinder into trying to scribble out of the ole again', ' the milkman came as usual', ' whos there he said also whispering standing under the window and peering up', ' not stopping to look again i reached the horses head hard round the right and another moment the dark heart had healed over upon the horse the shafts smashed toyzily and i was flung sideways and fell heavily into a shallow bowl of water', ' my god he said as i true men', ' so let traveling off the barrow fannels', ' they were all too as interestedly engaged to talk to us as we passed', ' the majority of people in london do not grade sunday papers', ' for a moment perhaps a hue did not clearly understand how i came there', ' where the road crossed narrow and black between the high peaks the crown cham and a desperate struggle occurred', ' i looked again at the martian and saw if he was now moving caseward along the durbank with a swift rolling martian', ' were the powder bills at unslow ready as a stern', ' i broached the mats surprised the size and whoresome in the shape the sposed meteorites around it pour less completely', ' only this morning i have achieved an early celebration', ' this was the second cylinder', ' it seemed there were a few people alive there fronting for the most part and many part and scalded', ' after that no body of men were standing into them so of us was the enterprise', ' and those who are outgill leave were not there', ' one of our shiplies cracked as if a shotten had hidden flew and beasts of it came clattering out the tiles of a deheap of broken bright frackiness upon the flowerbed by my study window', ' not far from me was a little oneroomed squatters out of wood surrounded by a batch of potato card', ' then it was and then only that he realized something of the full power and terror of these monsters', ' save that and unknown hellavent giving a group of four lined in the below of the spectrum as concerned we are still entirely ignorant of the nature of this substance', ' at that i gripped my wifes arm and without ceremony ran her out into the road', ' the occasional felling of the martians accesed they took up their bishes and the huge crescent upon their cylinders and absolute silent', ' what has it in the obscurity i could see he made a gesture of the spell', ' there was no sensible entrances from one seat of mine to the other', ' as i did so a second report followed and a main projectile hurled overhead towards a slow', ' the night was warm and still and a little oppressive the sound of guns continued intermittently and after midnight there seemed to me sheet lightning in the south', ' attended that night fourhanded systilagram describing in the gradual unscrewing of the shot was just to be a kenhard and deceiving paper after wiring for all the decay from him and receiving a reply the minimistic guild decided not to print a special edition', ' i cut up almost mechanically and went into the tidy room and drank some whiskey and then i was moved to change my clothes', ' london in danger of suffocation', ' they became fillers of bluntsofts smoke at the first touch of ray', ' talk about finchers of men fighters of fish it is this time', ' the respectable inhabitants of the place mint and cough and moat costumes the wye from pritly tres were backing riverside lovers energetically helping shintred excited and for the most part highly delighted at this astonishing variation of their settee experiences', ' i did not clearly remember the arrival of the curate so that probably i knows', ' there is one thing i said to allay the fears i had around they are the most sluckest things i ever saw crawl', ' we cut leatherhead without misfission about nine oclock and the horse had an hours rest where i took its supper with my cousins and cemented my wife to their care', ' with that i head', ' he returned and tried and faked the symptom delivered his attention he used examination notes', ' there was some movie from those more thruthless and extendable souls to whom a crowd has always adaptation for noise and horseplay', ' within vidk placed cveyn went to and fro over this glow about the little fears street and the deterrence of their empire over patr', ' he was a bit seeing sighted in the news and in the excess of its failings invited me up to take a turn with him that night in a scrutiny of the red planet', ' in the center stringing in the skin of our ultiplated earth like a poisoned art was this cylinder', ' but merely his nake had been broken', ' i thought through that the tit intrepidim had looked up at lee', ' here it was either hindering or escaping at the rim with the thin distilling sound', ' it had occurred towards midnight of the 12th and the spectroscope to which she had once resorted made a mass of flaming gas chiefly iced in moving with an enormous velocity towards this earth', ' so i fluffed out my house and copawn', ' it seemed to me that the bit had been enlarged and ever and again puffs of vivid green favour streamed up and out of it towards the brightening dawn streamed up whirled broke and vanished', ' but so little it was so still pretty warm a pinstead of light', ' i screamed loud and scalded how blinded agonized i staggered through the leaping as i went towards the shore', ' i struggled my feet at last ended crouching and made use of every chance of cover i made around for this', ' we went down the lane by the body of the man in black sudden now from the overnight hill and broke into the woods at the foot of the hill', ' this fitted with the vapor banks near us as we saw in the starlight from the window of a deserted house in upper halliford with her wee return', ' why the shots ceased after the death when on earth was attempted to explain', ' there was a noise of busyness flowing as works and the electric claims were all alight', ' i began walking the fore as a big curve sticking some point advantage and continually all looking into the standeaves that hid these new combusted wearers', ' he was death in one year', ' it was the head of the chopped man who had fallen in but showing as the little black object against the hot was turned to stir', ' but the common impulse we grabbered higher', ' they stop at the square as if undistided which way to date and finally turn eastward along the strand', ' the nature of the accident he could not ascertain indeed the railway authorities do not clearly know at that time', ' in addition ogilith wire to the usteromablik she and routes and creates observatory as its three', ' the thought the confined creature was though dreadful to him that he forgot the heat and went forth to the cylinder to now turn', ' what you mean a gun', ' and the rest of the world the stream of life still flowed as it had flowed for immemorial years', ' he was passed by a number of cyclists some horsemen and two rotor cars', ' i talked to her reassuringly bringing out that the martians were tied to the pit by sheer evidence and at the utmost could but crawl a little out of it which answered only in monosyllables']\\n\",\n",
       " 'wer: 0.27696012844527695\\n']"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lines"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "f4c0a1bd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'transduction_model.py --output_directory=/oak/stanford/projects/babelfish/magneto/GaddyPaper/output/efficiency_sweeps/S4/1 --epochs 80 --data_size_fraction=1 --hifigan_checkpoint=/oak/stanford/projects/babelfish/magneto/GaddyPaper/pretrained_models/hifigan_finetuned/checkpoint --S4=1 --model_size=700 --num_layers=8 --dropout=0.1 --learning_rate=1e-2'"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "' '.join(['transduction_model.py', '--output_directory=/oak/stanford/projects/babelfish/magneto/GaddyPaper/output/efficiency_sweeps/S4/1', '--epochs', '80', '--data_size_fraction=1', '--hifigan_checkpoint=/oak/stanford/projects/babelfish/magneto/GaddyPaper/pretrained_models/hifigan_finetuned/checkpoint', '--S4=1', '--model_size=700', '--num_layers=8', '--dropout=0.1', '--learning_rate=1e-2'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "c18511d0",
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'glob' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m/tmp/ipykernel_23718/2662821871.py\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[0mdata_dir\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m'/oak/stanford/projects/babelfish/magneto/GaddyPaper/output/EMGToText/HP_sweeps/S4/'\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0mdata_dir\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m'/oak/stanford/projects/babelfish/magneto/GaddyPaper/output/EMGToVoice/HP_sweeps/S4/'\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 4\u001b[0;31m \u001b[0msweeps\u001b[0m   \u001b[0;34m=\u001b[0m \u001b[0mglob\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mos\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpath\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mjoin\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdata_dir\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mparam\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'*'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      5\u001b[0m \u001b[0;31m#sweeps   = glob(os.path.join(data_dir, param, '*'))\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      6\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'glob' is not defined"
     ]
    }
   ],
   "source": [
    "param    = 'model_size'\n",
    "data_dir = '/oak/stanford/projects/babelfish/magneto/GaddyPaper/output/EMGToText/HP_sweeps/S4/'\n",
    "data_dir = '/oak/stanford/projects/babelfish/magneto/GaddyPaper/output/EMGToVoice/HP_sweeps/S4/'\n",
    "sweeps   = glob(os.path.join(data_dir, param, '*'))\n",
    "#sweeps   = glob(os.path.join(data_dir, param, '*'))\n",
    "\n",
    "scores, wers_dict = getSweepData(sweeps, param)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 187,
   "id": "1b971793",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1008x360 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from matplotlib import cm\n",
    "\n",
    "plt.figure(figsize=(14, 5))\n",
    "\n",
    "\n",
    "cmap = cm.get_cmap('plasma', 8)    # 11 discrete colors\n",
    "\n",
    "plt.subplot(1, 2, 1)\n",
    "\n",
    "for i, key in enumerate(sorted(wers_dict.keys())):\n",
    "    plt.plot(wers_dict[key], c = cmap(i), label = key)\n",
    "plt.ylim([20, None])\n",
    "plt.axhline(27, linestyle = '--', c = 'k')\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('WER')\n",
    "plt.legend()\n",
    "\n",
    "\n",
    "plt.subplot(1, 2, 2)\n",
    "\n",
    "sns.set(font_scale=1)\n",
    "sns.set_style(\"whitegrid\")\n",
    "sns.set_style({'axes.linewidth': 2, 'axes.edgecolor':'black'})\n",
    "\n",
    "sns.scatterplot(scores, x = param, y = 'WER', s=80)\n",
    "ax = plt.gca()\n",
    "ax.spines['top'].set_visible(False)\n",
    "ax.spines['right'].set_visible(False)\n",
    "plt.xlabel(param)\n",
    "#plt.xscale('log', base =10)\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "f5ed65e1",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/tmp/ipykernel_61569/3343985958.py:14: FutureWarning: The frame.append method is deprecated and will be removed from pandas in a future version. Use pandas.concat instead.\n",
      "  scores = scores.append([{param : val,\n",
      "/tmp/ipykernel_61569/3343985958.py:14: FutureWarning: The frame.append method is deprecated and will be removed from pandas in a future version. Use pandas.concat instead.\n",
      "  scores = scores.append([{param : val,\n",
      "/tmp/ipykernel_61569/3343985958.py:14: FutureWarning: The frame.append method is deprecated and will be removed from pandas in a future version. Use pandas.concat instead.\n",
      "  scores = scores.append([{param : val,\n",
      "/tmp/ipykernel_61569/3343985958.py:14: FutureWarning: The frame.append method is deprecated and will be removed from pandas in a future version. Use pandas.concat instead.\n",
      "  scores = scores.append([{param : val,\n",
      "/tmp/ipykernel_61569/3343985958.py:14: FutureWarning: The frame.append method is deprecated and will be removed from pandas in a future version. Use pandas.concat instead.\n",
      "  scores = scores.append([{param : val,\n",
      "/tmp/ipykernel_61569/3343985958.py:14: FutureWarning: The frame.append method is deprecated and will be removed from pandas in a future version. Use pandas.concat instead.\n",
      "  scores = scores.append([{param : val,\n",
      "/tmp/ipykernel_61569/3343985958.py:14: FutureWarning: The frame.append method is deprecated and will be removed from pandas in a future version. Use pandas.concat instead.\n",
      "  scores = scores.append([{param : val,\n",
      "/tmp/ipykernel_61569/3343985958.py:14: FutureWarning: The frame.append method is deprecated and will be removed from pandas in a future version. Use pandas.concat instead.\n",
      "  scores = scores.append([{param : val,\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 504x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(7, 5))\n",
    "\n",
    "param    = 'model_size'\n",
    "data_dir = '/oak/stanford/projects/babelfish/magneto/GaddyPaper/output/EMGToText/HP_sweeps/S4/'\n",
    "\n",
    "for layers in ['_8layers', '_9layers']:\n",
    "    sweeps   = glob(os.path.join(data_dir, param + '_1000Hz_longseq' + layers, '*'))\n",
    "    scores, wers_dict = getSweepData(sweeps, param)\n",
    "    \n",
    "    for i, key in enumerate(sorted(wers_dict.keys())):\n",
    "        plt.plot(wers_dict[key], label = str(int(key)) + layers)\n",
    "    plt.ylim([20, None])\n",
    "    plt.axhline(27, linestyle = '--', c = 'k')\n",
    "    plt.xlabel('Epochs')\n",
    "    plt.ylabel('WER')\n",
    "    plt.legend()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "fdad29d4",
   "metadata": {},
   "outputs": [],
   "source": [
    "import glob\n",
    "\n",
    "fname = glob.glob('/oak/stanford/projects/babelfish/magneto/GaddyPaper/processed_data/*/*/*.mat')[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "928ae8cc",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dict_keys(['__header__', '__version__', '__globals__', 'audio_features', 'emg', 'text', 'file_label', 'session_ids', 'book_location', 'silent', 'raw_emg', 'parallel_voiced_audio_features', 'parallel_voiced_emg', 'phonemes', 'audio_file'])"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import scipy\n",
    "\n",
    "scipy.io.loadmat(fname).keys()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "75a83cca",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "aa9579b8",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "id": "3f87fa78",
   "metadata": {},
   "outputs": [],
   "source": [
    "#lr_sched = torch.optim.lr_scheduler.MultiStepLR(optim, milestones=[125,150,175], gamma=.5)\n",
    "\n",
    "optim = torch.optim.AdamW(params = [torch.tensor(1)], lr = 1e-3)\n",
    "\n",
    "schedulers = [\n",
    "    torch.optim.lr_scheduler.MultiStepLR(optim, milestones=[150,175,200], gamma=.5),\n",
    "    torch.optim.lr_scheduler.OneCycleLR(optim, max_lr = 1e-3, epochs = 300,\n",
    "                                              steps_per_epoch = int(8055 / ( 2* 32)))\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "id": "75cc4e2c",
   "metadata": {},
   "outputs": [],
   "source": [
    "lrs_list = list()\n",
    "for idx, lr_sched in enumerate(schedulers):\n",
    "    lrs = list()\n",
    "    for i in range(300):\n",
    "        if idx == 0:\n",
    "            lr_sched.step()\n",
    "        else:\n",
    "            for j in range(int(8055 / ( 2* 32))):\n",
    "                lr_sched.step()\n",
    "        lrs.append(lr_sched.get_lr())\n",
    "    lrs_list.append(lrs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "id": "85d69eb0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "for lrs in lrs_list:\n",
    "    plt.plot(lrs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "f0886157",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<absl.flags._flagvalues.FlagHolder at 0x7f86715707f0>"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from absl import flags\n",
    "FLAGS = flags.FLAGS\n",
    "flags.DEFINE_boolean('debug', False, 'debug')\n",
    "flags.DEFINE_string('output_directory', 'output', 'where to save models and outputs')\n",
    "flags.DEFINE_integer('S4', 0, 'Toggle S4 model in place of transformer')\n",
    "flags.DEFINE_integer('batch_size', 32, 'training batch size')\n",
    "flags.DEFINE_float('learning_rate', 3e-4, 'learning rate')\n",
    "flags.DEFINE_integer('epochs', 200, 'training epochs')\n",
    "flags.DEFINE_integer('learning_rate_warmup', 1000, 'steps of linear warmup')\n",
    "flags.DEFINE_integer('learning_rate_patience', 5, 'learning rate decay patience')\n",
    "flags.DEFINE_string('start_training_from', None, 'start training from this model')\n",
    "flags.DEFINE_float('l2', 0, 'weight decay')\n",
    "flags.DEFINE_string('evaluate_saved', None, 'run evaluation on given model file')\n",
    "flags.DEFINE_string('lm_directory', '/oak/stanford/projects/babelfish/magneto/GaddyPaper/pretrained_models/librispeech_lm/', \n",
    "                    'Path to KenLM language model')\n",
    "flags.DEFINE_string('base_dir', '/oak/stanford/projects/babelfish/magneto/GaddyPaper/processed_data/',\n",
    "                    'path to processed EMG dataset')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "27ecd65f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['']"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "FLAGS([''])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0292d1d1",
   "metadata": {},
   "source": [
    "### Efficiency sweeps\n",
    "look at performance as a function of fraction of dataset retained."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1b60b49d",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "38e66c20",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fbc0e7b8",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "72cff08c",
   "metadata": {},
   "outputs": [
    {
     "ename": "ModuleNotFoundError",
     "evalue": "No module named 'jiwer'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mModuleNotFoundError\u001b[0m                       Traceback (most recent call last)",
      "\u001b[0;32m/tmp/ipykernel_8506/187747606.py\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0;32mimport\u001b[0m \u001b[0mnumpy\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mjiwer\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      2\u001b[0m \u001b[0;32mfrom\u001b[0m \u001b[0mIPython\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdisplay\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mMarkdown\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdisplay\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0;32mdef\u001b[0m \u001b[0mprintmd\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mstring\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m     \u001b[0mdisplay\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mMarkdown\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mstring\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mModuleNotFoundError\u001b[0m: No module named 'jiwer'"
     ]
    }
   ],
   "source": [
    "import numpy as np, jiwer\n",
    "from IPython.display import Markdown, display\n",
    "\n",
    "def printmd(string):\n",
    "    display(Markdown(string))\n",
    "\n",
    "examples = np.random.choice(np.arange(len(targets)), 5)\n",
    "\n",
    "for i, example_idx in enumerate(examples):\n",
    "    print('\\n')\n",
    "    printmd('**Target:** ' + targets[i])\n",
    "    printmd('**Prediction:**' + predictions[i])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "b26ce406",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.2659313725490196"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "removeBlanks = True\n",
    "\n",
    "if removeBlanks:\n",
    "    for i, (targ, pred) in enumerate(zip(targets, predictions)):\n",
    "        if targ == '':\n",
    "            del targets[i]\n",
    "            del predictions[i]\n",
    "\n",
    "transformation = jiwer.Compose([jiwer.RemovePunctuation(), jiwer.ToLowerCase()])\n",
    "targets     = transformation(targets)\n",
    "predictions = transformation(predictions)\n",
    "\n",
    "\n",
    "jiwer.wer(targets, predictions)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9579a1bf",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d5135b64",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f44cf789",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Magneto (3.9)",
   "language": "python",
   "name": "mag-3.9"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.15"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}