Datasets
Models
Downloads: 1
826 lines (825 with data), 37.5 kB
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# DREAMER Dominance FastGRNN 48_16"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Adapted from Microsoft's notebooks, available at https://github.com/microsoft/EdgeML authored by Dennis et al."
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [],
"source": [
"import pandas as pd\n",
"import numpy as np\n",
"from tabulate import tabulate\n",
"import os\n",
"import datetime as datetime\n",
"import pickle as pkl\n",
"from sklearn.model_selection import train_test_split\n",
"import pathlib\n",
"from os import mkdir"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [],
"source": [
"def loadData(dirname):\n",
" x_train = np.load(dirname + '/' + 'x_train.npy')\n",
" y_train = np.load(dirname + '/' + 'y_train.npy')\n",
" x_test = np.load(dirname + '/' + 'x_test.npy')\n",
" y_test = np.load(dirname + '/' + 'y_test.npy')\n",
" x_val = np.load(dirname + '/' + 'x_val.npy')\n",
" y_val = np.load(dirname + '/' + 'y_val.npy')\n",
" return x_train, y_train, x_test, y_test, x_val, y_val\n",
"def makeEMIData(subinstanceLen, subinstanceStride, sourceDir, outDir):\n",
" x_train, y_train, x_test, y_test, x_val, y_val = loadData(sourceDir)\n",
" x, y = bagData(x_train, y_train, subinstanceLen, subinstanceStride)\n",
" np.save(outDir + '/x_train.npy', x)\n",
" np.save(outDir + '/y_train.npy', y)\n",
" print('Num train %d' % len(x))\n",
" x, y = bagData(x_test, y_test, subinstanceLen, subinstanceStride)\n",
" np.save(outDir + '/x_test.npy', x)\n",
" np.save(outDir + '/y_test.npy', y)\n",
" print('Num test %d' % len(x))\n",
" x, y = bagData(x_val, y_val, subinstanceLen, subinstanceStride)\n",
" np.save(outDir + '/x_val.npy', x)\n",
" np.save(outDir + '/y_val.npy', y)\n",
" print('Num val %d' % len(x))\n",
"def bagData(X, Y, subinstanceLen, subinstanceStride):\n",
" numClass = 5\n",
" numSteps = 128\n",
" numFeats = 16\n",
" assert X.ndim == 3\n",
" assert X.shape[1] == numSteps\n",
" assert X.shape[2] == numFeats\n",
" assert subinstanceLen <= numSteps\n",
" assert subinstanceLen > 0\n",
" assert subinstanceStride <= numSteps\n",
" assert subinstanceStride >= 0\n",
" assert len(X) == len(Y)\n",
" assert Y.ndim == 2\n",
" assert Y.shape[1] == numClass\n",
" x_bagged = []\n",
" y_bagged = []\n",
" for i, point in enumerate(X[:, :, :]):\n",
" instanceList = []\n",
" start = 0\n",
" end = subinstanceLen\n",
" while True:\n",
" x = point[start:end, :]\n",
" if len(x) < subinstanceLen:\n",
" x_ = np.zeros([subinstanceLen, x.shape[1]])\n",
" x_[:len(x), :] = x[:, :]\n",
" x = x_\n",
" instanceList.append(x)\n",
" if end >= numSteps:\n",
" break\n",
" start += subinstanceStride\n",
" end += subinstanceStride\n",
" bag = np.array(instanceList)\n",
" numSubinstance = bag.shape[0]\n",
" label = Y[i]\n",
" label = np.argmax(label)\n",
" labelBag = np.zeros([numSubinstance, numClass])\n",
" labelBag[:, label] = 1\n",
" x_bagged.append(bag)\n",
" label = np.array(labelBag)\n",
" y_bagged.append(label)\n",
" return np.array(x_bagged), np.array(y_bagged)"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
"mkdir('/home/sf/data/DREAMER/Dominance/Fast_GRNN')"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Num train 61735\n",
"Num test 17149\n",
"Num val 6860\n"
]
}
],
"source": [
"subinstanceLen=48\n",
"subinstanceStride=16\n",
"extractedDir = '/home/sf/data/DREAMER/Dominance/'\n",
"mkdir('/home/sf/data/DREAMER/Dominance/Fast_GRNN/48_16')\n",
"rawDir = extractedDir + '/RAW'\n",
"sourceDir = rawDir\n",
"outDir = extractedDir + 'Fast_GRNN' '/%d_%d/' % (subinstanceLen, subinstanceStride)\n",
"makeEMIData(subinstanceLen, subinstanceStride, sourceDir, outDir)"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {
"ExecuteTime": {
"end_time": "2018-08-19T12:39:06.272261Z",
"start_time": "2018-08-19T12:39:05.330668Z"
}
},
"outputs": [],
"source": [
"from __future__ import print_function\n",
"import os\n",
"import sys\n",
"import tensorflow as tf\n",
"import numpy as np\n",
"os.environ['CUDA_VISIBLE_DEVICES'] = '0'\n",
"\n",
"# FastGRNN and FastRNN imports\n",
"from edgeml.graph.rnn import EMI_DataPipeline\n",
"from edgeml.graph.rnn import EMI_FastGRNN\n",
"from edgeml.graph.rnn import EMI_FastRNN\n",
"from edgeml.trainer.emirnnTrainer import EMI_Trainer, EMI_Driver\n",
"import edgeml.utils"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {
"ExecuteTime": {
"end_time": "2018-08-19T12:39:06.292205Z",
"start_time": "2018-08-19T12:39:06.274254Z"
}
},
"outputs": [],
"source": [
"# Network parameters for our FastGRNN + FC Layer\n",
"NUM_HIDDEN = 128\n",
"NUM_TIMESTEPS = 48\n",
"NUM_FEATS = 16\n",
"FORGET_BIAS = 1.0\n",
"NUM_OUTPUT = 5\n",
"USE_DROPOUT = False\n",
"KEEP_PROB = 0.9\n",
"\n",
"# Non-linearities can be chosen among \"tanh, sigmoid, relu, quantTanh, quantSigm\"\n",
"UPDATE_NL = \"quantTanh\"\n",
"GATE_NL = \"quantSigm\"\n",
"\n",
"# Ranks of Parameter matrices for low-rank parameterisation to compress models.\n",
"WRANK = 5\n",
"URANK = 6\n",
"\n",
"# For dataset API\n",
"PREFETCH_NUM = 5\n",
"BATCH_SIZE = 32\n",
"\n",
"# Number of epochs in *one iteration*\n",
"NUM_EPOCHS = 2\n",
"\n",
"# Number of iterations in *one round*. After each iteration,\n",
"# the model is dumped to disk. At the end of the current\n",
"# round, the best model among all the dumped models in the\n",
"# current round is picked up..\n",
"NUM_ITER = 4\n",
"\n",
"# A round consists of multiple training iterations and a belief\n",
"# update step using the best model from all of these iterations\n",
"NUM_ROUNDS = 10\n",
"\n",
"# A staging direcory to store models\n",
"MODEL_PREFIX = '/home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn'"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Loading Data"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"ExecuteTime": {
"end_time": "2018-08-19T12:39:06.410372Z",
"start_time": "2018-08-19T12:39:06.294014Z"
}
},
"outputs": [],
"source": [
"# Loading the data\n",
"path='/home/sf/data/DREAMER/Dominance/Fast_GRNN/48_16/'\n",
"x_train, y_train = np.load(path + 'x_train.npy'), np.load(path + 'y_train.npy')\n",
"x_test, y_test = np.load(path + 'x_test.npy'), np.load(path + 'y_test.npy')\n",
"x_val, y_val = np.load(path + 'x_val.npy'), np.load(path + 'y_val.npy')\n",
"\n",
"# BAG_TEST, BAG_TRAIN, BAG_VAL represent bag_level labels. These are used for the label update\n",
"# step of EMI/MI RNN\n",
"BAG_TEST = np.argmax(y_test[:, 0, :], axis=1)\n",
"BAG_TRAIN = np.argmax(y_train[:, 0, :], axis=1)\n",
"BAG_VAL = np.argmax(y_val[:, 0, :], axis=1)\n",
"NUM_SUBINSTANCE = x_train.shape[1]\n",
"print(\"x_train shape is:\", x_train.shape)\n",
"print(\"y_train shape is:\", y_train.shape)\n",
"print(\"x_test shape is:\", x_val.shape)\n",
"print(\"y_test shape is:\", y_val.shape)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Computation Graph"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {
"ExecuteTime": {
"end_time": "2018-08-19T12:39:06.653612Z",
"start_time": "2018-08-19T12:39:06.412290Z"
}
},
"outputs": [],
"source": [
"# Define the linear secondary classifier\n",
"def createExtendedGraph(self, baseOutput, *args, **kwargs):\n",
" W1 = tf.Variable(np.random.normal(size=[NUM_HIDDEN, NUM_OUTPUT]).astype('float32'), name='W1')\n",
" B1 = tf.Variable(np.random.normal(size=[NUM_OUTPUT]).astype('float32'), name='B1')\n",
" y_cap = tf.add(tf.tensordot(baseOutput, W1, axes=1), B1, name='y_cap_tata')\n",
" self.output = y_cap\n",
" self.graphCreated = True\n",
"\n",
"def restoreExtendedGraph(self, graph, *args, **kwargs):\n",
" y_cap = graph.get_tensor_by_name('y_cap_tata:0')\n",
" self.output = y_cap\n",
" self.graphCreated = True\n",
" \n",
"def feedDictFunc(self, keep_prob=None, inference=False, **kwargs):\n",
" if inference is False:\n",
" feedDict = {self._emiGraph.keep_prob: keep_prob}\n",
" else:\n",
" feedDict = {self._emiGraph.keep_prob: 1.0}\n",
" return feedDict\n",
"\n",
" \n",
"EMI_FastGRNN._createExtendedGraph = createExtendedGraph\n",
"EMI_FastGRNN._restoreExtendedGraph = restoreExtendedGraph\n",
"if USE_DROPOUT is True:\n",
" EMI_FastGRNN.feedDictFunc = feedDictFunc"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {
"ExecuteTime": {
"end_time": "2018-08-19T12:39:06.701740Z",
"start_time": "2018-08-19T12:39:06.655328Z"
}
},
"outputs": [],
"source": [
"inputPipeline = EMI_DataPipeline(NUM_SUBINSTANCE, NUM_TIMESTEPS, NUM_FEATS, NUM_OUTPUT)\n",
"emiFastGRNN = EMI_FastGRNN(NUM_SUBINSTANCE, NUM_HIDDEN, NUM_TIMESTEPS, NUM_FEATS, wRank=WRANK, uRank=URANK, \n",
" gate_non_linearity=GATE_NL, update_non_linearity=UPDATE_NL, useDropout=USE_DROPOUT)\n",
"emiTrainer = EMI_Trainer(NUM_TIMESTEPS, NUM_OUTPUT, lossType='xentropy')"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"x_train shape is: (61735, 6, 48, 16)\n",
"y_train shape is: (61735, 6, 5)\n",
"x_test shape is: (6860, 6, 48, 16)\n",
"y_test shape is: (6860, 6, 5)\n"
]
}
],
"source": [
"print(\"x_train shape is:\", x_train.shape)\n",
"print(\"y_train shape is:\", y_train.shape)\n",
"print(\"x_test shape is:\", x_val.shape)\n",
"print(\"y_test shape is:\", y_val.shape)"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {
"ExecuteTime": {
"end_time": "2018-08-19T12:39:14.187456Z",
"start_time": "2018-08-19T12:39:06.703481Z"
},
"scrolled": true
},
"outputs": [],
"source": [
"tf.reset_default_graph()\n",
"g1 = tf.Graph() \n",
"with g1.as_default():\n",
" # Obtain the iterators to each batch of the data\n",
" x_batch, y_batch = inputPipeline()\n",
" # Create the forward computation graph based on the iterators\n",
" y_cap = emiFastGRNN(x_batch)\n",
" # Create loss graphs and training routines\n",
" emiTrainer(y_cap, y_batch)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# EMI Driver"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {
"ExecuteTime": {
"end_time": "2018-08-19T12:51:45.803360Z",
"start_time": "2018-08-19T12:39:14.189648Z"
},
"scrolled": true
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Update policy: top-k\n",
"Training with MI-RNN loss for 5 rounds\n",
"Round: 0\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02630 Acc 0.47396 | Val acc 0.39242 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1000\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02545 Acc 0.47917 | Val acc 0.40262 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1001\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02509 Acc 0.47917 | Val acc 0.41399 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1002\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02476 Acc 0.53125 | Val acc 0.42274 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1003\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1003\n",
"Round: 1\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02451 Acc 0.55208 | Val acc 0.43921 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1004\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02395 Acc 0.56771 | Val acc 0.46487 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1005\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02386 Acc 0.57292 | Val acc 0.46778 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1006\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02396 Acc 0.58854 | Val acc 0.47755 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1007\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1007\n",
"Round: 2\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02389 Acc 0.60417 | Val acc 0.48294 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1008\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02375 Acc 0.58333 | Val acc 0.49213 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1009\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02292 Acc 0.61979 | Val acc 0.50029 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1010\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02309 Acc 0.58333 | Val acc 0.50131 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1011\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1011\n",
"Round: 3\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02265 Acc 0.60417 | Val acc 0.50102 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1012\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02233 Acc 0.61458 | Val acc 0.50073 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1013\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02249 Acc 0.59896 | Val acc 0.51662 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1014\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02244 Acc 0.59375 | Val acc 0.51618 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1015\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1014\n",
"Round: 4\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02244 Acc 0.59375 | Val acc 0.51618 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1016\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02242 Acc 0.55208 | Val acc 0.51531 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1017\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02217 Acc 0.59896 | Val acc 0.52172 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1018\n",
"Epoch 1 Batch 1925 ( 3855) Loss 0.02183 Acc 0.61458 | Val acc 0.52041 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1019\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1018\n",
"Round: 5\n",
"Switching to EMI-Loss function\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.15738 Acc 0.53125 | Val acc 0.50058 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1020\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.16199 Acc 0.54167 | Val acc 0.50875 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1021\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.15353 Acc 0.54688 | Val acc 0.51545 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1022\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.13855 Acc 0.55208 | Val acc 0.52157 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1023\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1023\n",
"Round: 6\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.12964 Acc 0.58854 | Val acc 0.52187 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1024\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.11410 Acc 0.59375 | Val acc 0.52289 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1025\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.11220 Acc 0.59896 | Val acc 0.52172 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1026\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.10395 Acc 0.60417 | Val acc 0.52318 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1027\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1027\n",
"Round: 7\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.11053 Acc 0.60417 | Val acc 0.52536 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1028\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.12034 Acc 0.57812 | Val acc 0.53120 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1029\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.12177 Acc 0.57292 | Val acc 0.52959 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1030\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.12257 Acc 0.58854 | Val acc 0.53265 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1031\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1031\n",
"Round: 8\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.14379 Acc 0.58333 | Val acc 0.52857 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1032\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.13761 Acc 0.61458 | Val acc 0.53207 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1033\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.13618 Acc 0.59375 | Val acc 0.53265 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1034\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.12991 Acc 0.59375 | Val acc 0.53222 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1035\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1034\n",
"Round: 9\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.12991 Acc 0.59375 | Val acc 0.53222 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1036\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.10665 Acc 0.60938 | Val acc 0.52901 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1037\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.10800 Acc 0.60938 | Val acc 0.53207 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1038\n",
"Epoch 1 Batch 1925 ( 3855) Loss 1.08893 Acc 0.61458 | Val acc 0.53265 | Model saved to /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn, global_step 1039\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1039\n"
]
}
],
"source": [
"with g1.as_default():\n",
" emiDriver = EMI_Driver(inputPipeline, emiFastGRNN, emiTrainer)\n",
"\n",
"emiDriver.initializeSession(g1)\n",
"y_updated, modelStats = emiDriver.run(numClasses=NUM_OUTPUT, x_train=x_train,\n",
" y_train=y_train, bag_train=BAG_TRAIN,\n",
" x_val=x_val, y_val=y_val, bag_val=BAG_VAL,\n",
" numIter=NUM_ITER, keep_prob=KEEP_PROB,\n",
" numRounds=NUM_ROUNDS, batchSize=BATCH_SIZE,\n",
" numEpochs=NUM_EPOCHS, modelPrefix=MODEL_PREFIX,\n",
" fracEMI=0.5, updatePolicy='top-k', k=1)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Evaluating the trained model"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {
"ExecuteTime": {
"end_time": "2018-08-19T12:51:45.832728Z",
"start_time": "2018-08-19T12:51:45.805984Z"
}
},
"outputs": [],
"source": [
"# Early Prediction Policy: We make an early prediction based on the predicted classes\n",
"# probability. If the predicted class probability > minProb at some step, we make\n",
"# a prediction at that step.\n",
"def earlyPolicy_minProb(instanceOut, minProb, **kwargs):\n",
" assert instanceOut.ndim == 2\n",
" classes = np.argmax(instanceOut, axis=1)\n",
" prob = np.max(instanceOut, axis=1)\n",
" index = np.where(prob >= minProb)[0]\n",
" if len(index) == 0:\n",
" assert (len(instanceOut) - 1) == (len(classes) - 1)\n",
" return classes[-1], len(instanceOut) - 1\n",
" index = index[0]\n",
" return classes[index], index\n",
"\n",
"def getEarlySaving(predictionStep, numTimeSteps, returnTotal=False):\n",
" predictionStep = predictionStep + 1\n",
" predictionStep = np.reshape(predictionStep, -1)\n",
" totalSteps = np.sum(predictionStep)\n",
" maxSteps = len(predictionStep) * numTimeSteps\n",
" savings = 1.0 - (totalSteps / maxSteps)\n",
" if returnTotal:\n",
" return savings, totalSteps\n",
" return savings"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {
"ExecuteTime": {
"end_time": "2018-08-19T12:51:46.210240Z",
"start_time": "2018-08-19T12:51:45.834534Z"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Test Accuracy (k = 2): 0.528194\n",
"\n",
"Total Savings: 0.629757\n"
]
}
],
"source": [
"k = 2\n",
"ORIGINAL_NUM_TIMESTEPS = 128\n",
"predictions, predictionStep = emiDriver.getInstancePredictions(x_test, y_test, earlyPolicy_minProb, minProb=0.99)\n",
"bagPredictions = emiDriver.getBagPredictions(predictions, minSubsequenceLen=k, numClass=NUM_OUTPUT)\n",
"print('Test Accuracy (k = %d): %f\\n' % (k, np.mean((bagPredictions == BAG_TEST).astype(int))))\n",
"mi_savings = (1 - NUM_TIMESTEPS / ORIGINAL_NUM_TIMESTEPS)\n",
"emi_savings = getEarlySaving(predictionStep, NUM_TIMESTEPS)\n",
"total_savings = mi_savings + (1 - mi_savings) * emi_savings\n",
"print(\"Total Savings: %f\" % total_savings)"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {
"ExecuteTime": {
"end_time": "2018-08-19T12:51:46.677691Z",
"start_time": "2018-08-19T12:51:46.212285Z"
},
"scrolled": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
" len acc macro-fsc macro-pre macro-rec micro-fsc micro-pre \\\n",
"0 1 0.529535 0.448461 0.626394 0.424688 0.529535 0.529535 \n",
"1 2 0.528194 0.454806 0.504050 0.435102 0.528194 0.528194 \n",
"2 3 0.481777 0.417933 0.478811 0.430978 0.481777 0.481777 \n",
"3 4 0.396058 0.375566 0.522220 0.401936 0.396058 0.396058 \n",
"4 5 0.336055 0.343254 0.570104 0.381710 0.336055 0.336055 \n",
"5 6 0.292437 0.315475 0.611050 0.365957 0.292437 0.292437 \n",
"\n",
" micro-rec \n",
"0 0.529535 \n",
"1 0.528194 \n",
"2 0.481777 \n",
"3 0.396058 \n",
"4 0.336055 \n",
"5 0.292437 \n",
"Max accuracy 0.529535 at subsequencelength 1\n",
"Max micro-f 0.529535 at subsequencelength 1\n",
"Micro-precision 0.529535 at subsequencelength 1\n",
"Micro-recall 0.529535 at subsequencelength 1\n",
"Max macro-f 0.454806 at subsequencelength 2\n",
"macro-precision 0.504050 at subsequencelength 2\n",
"macro-recall 0.435102 at subsequencelength 2\n"
]
}
],
"source": [
"# A slightly more detailed analysis method is provided. \n",
"df = emiDriver.analyseModel(predictions, BAG_TEST, NUM_SUBINSTANCE, NUM_OUTPUT)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Picking the best model"
]
},
{
"cell_type": "code",
"execution_count": 18,
"metadata": {
"ExecuteTime": {
"end_time": "2018-08-19T13:06:04.024660Z",
"start_time": "2018-08-19T13:04:47.045787Z"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1003\n",
"Round: 0, Validation accuracy: 0.4227, Test Accuracy (k = 2): 0.425622, Additional savings: 0.000027\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1007\n",
"Round: 1, Validation accuracy: 0.4776, Test Accuracy (k = 2): 0.477346, Additional savings: 0.001708\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1011\n",
"Round: 2, Validation accuracy: 0.5013, Test Accuracy (k = 2): 0.499738, Additional savings: 0.003691\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1014\n",
"Round: 3, Validation accuracy: 0.5166, Test Accuracy (k = 2): 0.507377, Additional savings: 0.004199\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1018\n",
"Round: 4, Validation accuracy: 0.5217, Test Accuracy (k = 2): 0.516182, Additional savings: 0.004893\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1023\n",
"Round: 5, Validation accuracy: 0.5216, Test Accuracy (k = 2): 0.504111, Additional savings: 0.009568\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1027\n",
"Round: 6, Validation accuracy: 0.5232, Test Accuracy (k = 2): 0.515832, Additional savings: 0.011143\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1031\n",
"Round: 7, Validation accuracy: 0.5327, Test Accuracy (k = 2): 0.519331, Additional savings: 0.012006\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1034\n",
"Round: 8, Validation accuracy: 0.5327, Test Accuracy (k = 2): 0.520555, Additional savings: 0.012196\n",
"INFO:tensorflow:Restoring parameters from /home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn-1039\n",
"Round: 9, Validation accuracy: 0.5327, Test Accuracy (k = 2): 0.528194, Additional savings: 0.012686\n"
]
}
],
"source": [
"devnull = open(os.devnull, 'r')\n",
"for val in modelStats:\n",
" round_, acc, modelPrefix, globalStep = val\n",
" emiDriver.loadSavedGraphToNewSession(modelPrefix, globalStep, redirFile=devnull)\n",
" predictions, predictionStep = emiDriver.getInstancePredictions(x_test, y_test, earlyPolicy_minProb,\n",
" minProb=0.99, keep_prob=1.0)\n",
" \n",
" bagPredictions = emiDriver.getBagPredictions(predictions, minSubsequenceLen=k, numClass=NUM_OUTPUT)\n",
" print(\"Round: %2d, Validation accuracy: %.4f\" % (round_, acc), end='')\n",
" print(', Test Accuracy (k = %d): %f, ' % (k, np.mean((bagPredictions == BAG_TEST).astype(int))), end='')\n",
" print('Additional savings: %f' % getEarlySaving(predictionStep, NUM_TIMESTEPS)) "
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {},
"outputs": [],
"source": [
"# Network parameters for our FastGRNN + FC Layer\n",
"NUM_HIDDEN = 128\n",
"NUM_TIMESTEPS = 48\n",
"NUM_FEATS = 16\n",
"FORGET_BIAS = 1.0\n",
"NUM_OUTPUT = 5\n",
"USE_DROPOUT = False\n",
"KEEP_PROB = 0.9\n",
"\n",
"# Non-linearities can be chosen among \"tanh, sigmoid, relu, quantTanh, quantSigm\"\n",
"UPDATE_NL = \"quantTanh\"\n",
"GATE_NL = \"quantSigm\"\n",
"\n",
"# Ranks of Parameter matrices for low-rank parameterisation to compress models.\n",
"WRANK = 5\n",
"URANK = 6\n",
"\n",
"# For dataset API\n",
"PREFETCH_NUM = 5\n",
"BATCH_SIZE = 32\n",
"\n",
"# Number of epochs in *one iteration*\n",
"NUM_EPOCHS = 2\n",
"# Number of iterations in *one round*. After each iteration,\n",
"# the model is dumped to disk. At the end of the current\n",
"# round, the best model among all the dumped models in the\n",
"# current round is picked up..\n",
"NUM_ITER = 4\n",
"# A round consists of multiple training iterations and a belief\n",
"# update step using the best model from all of these iterations\n",
"NUM_ROUNDS = 10\n",
"\n",
"# A staging direcory to store models\n",
"MODEL_PREFIX = '/home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn'"
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {},
"outputs": [],
"source": [
"params = {\n",
" \"NUM_HIDDEN\" : 128,\n",
" \"NUM_TIMESTEPS\" : 48, #subinstance length.\n",
" \"ORIGINAL_NUM_TIMESTEPS\" : 128,\n",
" \"NUM_FEATS\" : 16,\n",
" \"FORGET_BIAS\" : 1.0,\n",
" \"UPDATE_NL\" : \"quantTanh\",\n",
" \"GATE_NL\" : \"quantSigm\",\n",
" \"NUM_OUTPUT\" : 5,\n",
" \"WRANK\" : 5,\n",
" \"URANK\" : 6,\n",
" \"USE_DROPOUT\" : 0, # '1' -> True. '0' -> False\n",
" \"KEEP_PROB\" : 0.9,\n",
" \"PREFETCH_NUM\" : 5,\n",
" \"BATCH_SIZE\" : 32,\n",
" \"NUM_EPOCHS\" : 2,\n",
" \"NUM_ITER\" : 4,\n",
" \"NUM_ROUNDS\" : 10,\n",
" \"MODEL_PREFIX\" : '/home/sf/data/DREAMER/Dominance/48_16/models/Fast-GRNN/model-fgrnn'\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": 21,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
" len acc macro-fsc macro-pre macro-rec micro-fsc micro-pre \\\n",
"0 1 0.529535 0.448461 0.626394 0.424688 0.529535 0.529535 \n",
"1 2 0.528194 0.454806 0.504050 0.435102 0.528194 0.528194 \n",
"2 3 0.481777 0.417933 0.478811 0.430978 0.481777 0.481777 \n",
"3 4 0.396058 0.375566 0.522220 0.401936 0.396058 0.396058 \n",
"4 5 0.336055 0.343254 0.570104 0.381710 0.336055 0.336055 \n",
"5 6 0.292437 0.315475 0.611050 0.365957 0.292437 0.292437 \n",
"\n",
" micro-rec \n",
"0 0.529535 \n",
"1 0.528194 \n",
"2 0.481777 \n",
"3 0.396058 \n",
"4 0.336055 \n",
"5 0.292437 \n",
"Max accuracy 0.529535 at subsequencelength 1\n",
"Max micro-f 0.529535 at subsequencelength 1\n",
"Micro-precision 0.529535 at subsequencelength 1\n",
"Micro-recall 0.529535 at subsequencelength 1\n",
"Max macro-f 0.454806 at subsequencelength 2\n",
"macro-precision 0.504050 at subsequencelength 2\n",
"macro-recall 0.435102 at subsequencelength 2\n",
"+----+-------+----------+-------------+-------------+-------------+-------------+-------------+-------------+\n",
"| | len | acc | macro-fsc | macro-pre | macro-rec | micro-fsc | micro-pre | micro-rec |\n",
"+====+=======+==========+=============+=============+=============+=============+=============+=============+\n",
"| 0 | 1 | 0.529535 | 0.448461 | 0.626394 | 0.424688 | 0.529535 | 0.529535 | 0.529535 |\n",
"+----+-------+----------+-------------+-------------+-------------+-------------+-------------+-------------+\n",
"| 1 | 2 | 0.528194 | 0.454806 | 0.50405 | 0.435102 | 0.528194 | 0.528194 | 0.528194 |\n",
"+----+-------+----------+-------------+-------------+-------------+-------------+-------------+-------------+\n",
"| 2 | 3 | 0.481777 | 0.417933 | 0.478811 | 0.430978 | 0.481777 | 0.481777 | 0.481777 |\n",
"+----+-------+----------+-------------+-------------+-------------+-------------+-------------+-------------+\n",
"| 3 | 4 | 0.396058 | 0.375566 | 0.52222 | 0.401936 | 0.396058 | 0.396058 | 0.396058 |\n",
"+----+-------+----------+-------------+-------------+-------------+-------------+-------------+-------------+\n",
"| 4 | 5 | 0.336055 | 0.343254 | 0.570104 | 0.38171 | 0.336055 | 0.336055 | 0.336055 |\n",
"+----+-------+----------+-------------+-------------+-------------+-------------+-------------+-------------+\n",
"| 5 | 6 | 0.292437 | 0.315475 | 0.61105 | 0.365957 | 0.292437 | 0.292437 | 0.292437 |\n",
"+----+-------+----------+-------------+-------------+-------------+-------------+-------------+-------------+\n"
]
}
],
"source": [
"fgrnn_dict = {**params}\n",
"fgrnn_dict[\"k\"] = k\n",
"fgrnn_dict[\"accuracy\"] = np.mean((bagPredictions == BAG_TEST).astype(int))\n",
"fgrnn_dict[\"total_savings\"] = getEarlySaving(predictionStep, NUM_TIMESTEPS)\n",
"fgrnn_dict[\"y_test\"] = BAG_TEST\n",
"fgrnn_dict[\"y_pred\"] = bagPredictions\n",
"\n",
"# A slightly more detailed analysis method is provided. \n",
"df = emiDriver.analyseModel(predictions, BAG_TEST, NUM_SUBINSTANCE, NUM_OUTPUT)\n",
"print (tabulate(df, headers=list(df.columns), tablefmt='grid'))"
]
},
{
"cell_type": "code",
"execution_count": 22,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Results for this run have been saved at /home/sf/data/DREAMER/Dominance/Fast_GRNN/ .\n"
]
}
],
"source": [
"dirname = \"/home/sf/data/DREAMER/Dominance/Fast_GRNN/\"\n",
"pathlib.Path(dirname).mkdir(parents=True, exist_ok=True)\n",
"print (\"Results for this run have been saved at\" , dirname, \".\")\n",
"\n",
"now = datetime.datetime.now()\n",
"filename = list((str(now.year),\"-\",str(now.month),\"-\",str(now.day),\"|\",str(now.hour),\"-\",str(now.minute)))\n",
"filename = ''.join(filename)\n",
"\n",
"#Save the dictionary containing the params and the results.\n",
"pkl.dump(fgrnn_dict,open(dirname + filename + \".pkl\",mode='wb'))"
]
}
],
"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": "3.7.3"
}
},
"nbformat": 4,
"nbformat_minor": 2
}