intro-to-dltrainers.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Populating the interactive namespace from numpy and matplotlib\n"
     ]
    }
   ],
   "source": [
    "%pylab inline\n",
    "import numpy as np\n",
    "import scipy as sc\n",
    "import scipy.ndimage as ndi\n",
    "import pylab as pl\n",
    "import matplotlib as mpl\n",
    "from IPython import display\n",
    "from itertools import islice\n",
    "rc(\"image\", cmap=\"gray\")\n",
    "from dlinputs import tarrecords, gopen, utils, filters, sqlitedb\n",
    "import dltrainers as dlt\n",
    "import dltrainers.layers as dll\n",
    "import torch\n",
    "from torch import nn\n",
    "from torch.autograd import Variable"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Flex Models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Sequential(\n",
       "  (0): Flex:None\n",
       "  (1): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), dilation=(1, 1), ceil_mode=False)\n",
       "  (2): Flex:None\n",
       "  (3): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), dilation=(1, 1), ceil_mode=False)\n",
       "  (4): Flat\n",
       "  (5): Flex:None\n",
       "  (6): ReLU()\n",
       "  (7): Flex:None\n",
       "  (8): Sigmoid()\n",
       ")"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = nn.Sequential(\n",
    "    dll.Conv2d(64, 3),\n",
    "    nn.MaxPool2d(2),\n",
    "    dll.Conv2d(64, 3),\n",
    "    nn.MaxPool2d(2),\n",
    "    dll.Flat(),\n",
    "    dll.Linear(100),\n",
    "    nn.ReLU(),\n",
    "    dll.Linear(10),\n",
    "    nn.Sigmoid()\n",
    ")\n",
    "model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Sequential(\n",
       "  (0): Flex:Conv2d(1, 64, kernel_size=(3, 3), stride=(1, 1))\n",
       "  (1): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), dilation=(1, 1), ceil_mode=False)\n",
       "  (2): Flex:Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1))\n",
       "  (3): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), dilation=(1, 1), ceil_mode=False)\n",
       "  (4): Flat\n",
       "  (5): Flex:Linear(in_features=1600, out_features=100, bias=True)\n",
       "  (6): ReLU()\n",
       "  (7): Flex:Linear(in_features=100, out_features=10, bias=True)\n",
       "  (8): Sigmoid()\n",
       ")"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x = torch.randn(10, 1, 28, 28)\n",
    "model.forward(Variable(x))\n",
    "model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Trainers\n",
    "\n",
    "Trainers allow quick and simple training of models. Unlike other deep learning frameworks, trainers are just temporary scaffolds for training a model; that is, they are wrapped around a model, used for training, and then discarded, leaving only the trained model.\n",
    "\n",
    "In addition, trainers do not invert control; that is, you write a standard stochastic gradient descent training loop. This is not only less mysterious than burying the loop inside some framework, it also gives a great deal more flexibility in terms of training logic.\n",
    "\n",
    "Trainers also provide convenient logic for dynamic displays inside iPython notbooks (`trainer.display_loss()`)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "import dltrainers as dlt\n",
    "trainer = dlt.ImageClassifierTrainer(model)\n",
    "trainer.set_lr(1e-2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['image', 'inx', '__epoch__', 'cls']"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "source = sqlitedb.sqlitedb(\"testdata/sample.db\", epochs=5000)\n",
    "pipeline = filters.compose(\n",
    "    filters.map(image=utils.pilgray, cls=int),\n",
    "    filters.batched(17),\n",
    "    filters.sliced(0, 1000)\n",
    ")\n",
    "data = pipeline(source)\n",
    "data.next().keys()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "for sample in data:\n",
    "    inputs = expand_dims(sample[\"image\"], 3)\n",
    "    classes = sample[\"cls\"]\n",
    "    outputs, loss = trainer.train_batch(inputs, classes)\n",
    "    clf()\n",
    "    xscale('log'); yscale('log')\n",
    "    trainer.plot_loss()\n",
    "    dlt.update_display()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}