infer_andi.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import csv as csv\n",
    "from tensorflow.keras.models import load_model\n",
    "from utils import many_net_uhd"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In the example file we saw how to use the networks for trajectories of equal length. Now let's see how to use them for trajectories of different lengths as the ones provided in the AnDi Challenge"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Inference"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Importing the data, if you have not already done so, download the data [here](https://competitions.codalab.org/my/datasets/download/7ea12913-dfcf-4a50-9f5d-8bf9666e9bb4) and unzip it in this directory. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import csv\n",
    "trajs_from_files = csv.reader(open('challenge_for_scoring/task1.txt','r'), delimiter=';', \n",
    "                                        lineterminator='\\n',quoting=csv.QUOTE_NONNUMERIC)\n",
    "validation = [[],[],[]]\n",
    "for trajs in enumerate(trajs_from_files):\n",
    "    validation[int(trajs[1][0])-1].append(trajs[1][1:])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Inference"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The output of the inference networks is the inferred anomalous exponent for each trajectory. Trajectories must be of equal length"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 1d"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Importing the networks used for inference in 1d"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "centers_inf_1d = [25, 50, 65, 75, 125, 165, 225, \n",
    "                  325, 425, 525, 625, 725, 825, 925]\n",
    "\n",
    "meta_model_inf_1d = []\n",
    "for i in centers_inf_1d: \n",
    "    m = load_model('nets/inference_nets/1d/inference_1D_'+str(i)+'.h5')\n",
    "    \n",
    "    meta_model_inf_1d.append(m)\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Printing the block size required by each net"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for n in meta_model_inf_1d:\n",
    "        print(n.layers[0].input_shape[-1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "traj_set = validation[0]\n",
    "tot_tt=len(traj_set)\n",
    "count=0\n",
    "pred_inf_1d = []\n",
    "for traj in traj_set:\n",
    "    count=count+1\n",
    "    aa = np.asarray(traj).reshape(1,-1)\n",
    "    pr = many_net_uhd(nets = meta_model_inf_1d, traj_set = aa, centers = centers_inf_1d ,dim = 1, task =1)\n",
    "   # print(pr)\n",
    "    pred_inf_1d.append(pr.flatten())\n",
    "    print('traj',count,'/',tot_tt,end='\\r')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Inference in 2d"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "centers_inf_2d = [25, 65, 125, 225, 325, 425, 525, 725, 925]\n",
    "meta_model_inf_2d = []\n",
    "for i in centers_inf_2d: \n",
    "    m = load_model('nets/inference_nets/2d/inference_2D_'+str(i)+'.h5')\n",
    "    \n",
    "    meta_model_inf_2d.append(m)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "traj_set = validation[1]\n",
    "tot_tt=len(traj_set)\n",
    "count=0\n",
    "pred_inf_2d = []\n",
    "for traj in traj_set:\n",
    "    count=count+1\n",
    "    aa = np.asarray(traj).reshape(1,-1)\n",
    "    pr = many_net_uhd(nets = meta_model_inf_2d, traj_set = aa, centers = centers_inf_2d ,dim = 2, task =1)\n",
    "   # print(pr)\n",
    "    pred_inf_2d.append(pr.flatten())\n",
    "    print('traj',count,'/',tot_tt,end='\\r')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 3d"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "centers_inf_3d = [25, 65, 125, 225, 525, 925]\n",
    "meta_model_inf_3d = []\n",
    "for i in centers_inf_3d: \n",
    "    m = load_model('nets/inference_nets/3d/inference_3D_'+str(i)+'.h5')\n",
    "    \n",
    "    meta_model_inf_3d.append(m)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "traj_set = validation[2]\n",
    "tot_tt=len(traj_set)\n",
    "count=0\n",
    "pred_inf_3d = []\n",
    "for traj in traj_set:\n",
    "    count=count+1\n",
    "    aa = np.asarray(traj).reshape(1,-1)\n",
    "    pr = many_net_uhd(nets = meta_model_inf_3d, traj_set = aa, centers = centers_inf_3d ,dim = 3, task =1)\n",
    "   # print(pr)\n",
    "    pred_inf_3d.append(pr.flatten())\n",
    "    print('traj',count,'/',tot_tt,end='\\r')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}