Skip to content

Instantly share code, notes, and snippets.

@anonoz

anonoz/demo_sos_keras_own_dataset.ipynb

Created September 29, 2016 19:54
Show Gist options
  • Save anonoz/cf7421a29818dd5a843371ee23408cb1 to your computer and use it in GitHub Desktop.
Save anonoz/cf7421a29818dd5a843371ee23408cb1 to your computer and use it in GitHub Desktop.
Download ZIP
Display the source blob
Display the rendered blob
Raw
demo_sos_keras_own_dataset.ipynb
{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "aa8439c7-cf04-49a6-84f3-2e8af08aca7f"
}
},
"source": [
"# Sentiment Analysis using word vector and ConvNet in Keras"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "07039836-86fc-4567-b302-ef9265af8307"
}
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Using gpu device 0: GeForce GTX 1060 6GB (CNMeM is disabled, cuDNN 5105)\n",
"/home/anonoz/anaconda2/envs/tensorflow/lib/python2.7/site-packages/theano/sandbox/cuda/__init__.py:600: UserWarning: Your cuDNN version is more recent than the one Theano officially supports. If you see any problems, try updating Theano or downgrading cuDNN to version 5.\n",
" warnings.warn(warn)\n"
]
}
],
"source": [
"# CONFIGS\n",
"\n",
"# The word vector can be swapped with say GoogleNews 6B dataset\n",
"word_vector_bin_file = \"word2vec/model-0.bin\"\n",
"# word_vector_bin_file = \"word2vec/GoogleNews-vectors-negative300.bin\"\n",
"# word_vector_bin_file = \"model-0.bin\"\n",
"word_vector_dims = 100\n",
"\n",
"# in aclImdb, the longest review is 2470 words long\n",
"# Due to memory constraint, in this one I limit to 200 words\n",
"max_sentence_length = 200\n",
"\n",
"# Can easily swap with other datasets if you want\n",
"positive_review_txts = \"aclImdb/train/pos/*.txt\"\n",
"negative_review_txts = \"aclImdb/train/neg/*.txt\"\n",
"positive_review_vals = \"aclImdb/test/pos/*.txt\"\n",
"negative_review_vals = \"aclImdb/test/neg/*.txt\"\n",
"# positive_review_txts = \"polarity2/txt_sentoken/pos/*.txt\"\n",
"# negative_review_txts = \"polarity2/txt_sentoken/neg/*.txt\"\n",
"# positive_review_vals = \"polarity2/txt_sentoken/pos/*.txt\"\n",
"# negative_review_vals = \"polarity2/txt_sentoken/neg/*.txt\"\n",
"\n",
"# Test theano and graphics card\n",
"import theano.tensor as T\n",
"\n",
"# Notebook\n",
"%matplotlib inline\n",
"import matplotlib.pyplot as plt\n",
"import numpy as np"
]
},
{
"cell_type": "markdown",
"metadata": {
"collapsed": true,
"nbpresent": {
"id": "35ac46e5-404e-4d92-b9bd-073c346223a0"
}
},
"source": [
"## Step 1: Make X_train\n",
"\n",
"X_train data structure is a 3D array, consisting of reviews, words, and vectors:\n",
"```json\n",
"[\n",
" // a review\n",
" [\n",
" // a word, and its array of 100 vectors\n",
" [0.75, 0.64 ...],\n",
" ...\n",
" ], \n",
" ...\n",
"]\n",
"```\n",
"\n",
"### TODO:\n",
"1. Load all the reviews into memory\n",
"2. Normalize the text\n",
"3. Add words to vocab array to make word vector retrieval faster\n",
"3. Determine vocab size, max review length"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "bc104c55-9cb1-4d4c-a1bf-363277a69da8"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"omg this is soo funny \n",
"[u'omg', u'this', u'is', u'soo', u'funny']\n"
]
}
],
"source": [
"from bs4 import BeautifulSoup \n",
"import re\n",
"\n",
"def normalise_text(text):\n",
" # Step 0: Clean HTML tags\n",
" normalised_text = BeautifulSoup(text, \"html.parser\").get_text()\n",
" \n",
" # Step 1: Lower cap\n",
" normalised_text = normalised_text.lower()\n",
" \n",
" # Step 2: Leave only abc behind\n",
" normalised_text = re.sub(\"[^a-zA-Z ]\", \" \", normalised_text)\n",
" \n",
" # Step 3: \n",
" return normalised_text\n",
"\n",
"# Test normalise_text\n",
"print(normalise_text(\"OMG <br/><br />this is soo00.... funny!!!\"))\n",
"\n",
"def raw_to_array(review_text):\n",
" return normalise_text(review_text).split()\n",
"\n",
"# Test both fx above\n",
"print(raw_to_array(\"OMG <br/><br />this is soo00.... funny!!!\"))"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "789db32e-3bb2-49b4-8ecc-e8f9f0d67a7b"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"('highest word count: ', 2487)\n",
"[u'i', u'remember', u'watching', u'this', u'late', u'at', u'night', u'on', u'black', u'and', u'white', u'tv', u'long', u'before', u'a', u'live', u'action', u'version', u'was', u'so']\n",
"('mean of review length: ', 236.82848000000001)\n"
]
},
{
"data": {
"text/plain": [
"[<matplotlib.lines.Line2D at 0x7fc3c8190290>]"
]
},
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
},
{
"data": {
"image/png": "iVBORw0KGgoAAAANSUhEUgAAAYoAAAEACAYAAACtVTGuAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
"text/plain": [
"<matplotlib.figure.Figure at 0x7fc3d5633750>"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"import glob\n",
"\n",
"# highest word count shall be the convnet rows\n",
"highest_review_word_count = 0\n",
"positive_reviews = []\n",
"negative_reviews = []\n",
"positive_testrev = []\n",
"negative_testrev = []\n",
"\n",
"# to visualise review length\n",
"training_review_length = []\n",
"\n",
"for txt in glob.glob(positive_review_txts):\n",
" with (open(txt, 'r')) as f:\n",
" word_array = raw_to_array(f.read())\n",
" highest_review_word_count = max(highest_review_word_count, len(word_array))\n",
" positive_reviews.append(word_array)\n",
" training_review_length.append(len(word_array))\n",
"\n",
"for txt in glob.glob(negative_review_txts):\n",
" with (open(txt, 'r')) as f:\n",
" word_array = raw_to_array(f.read())\n",
" highest_review_word_count = max(highest_review_word_count, len(word_array))\n",
" negative_reviews.append(raw_to_array(f.read()))\n",
" training_review_length.append(len(word_array))\n",
" \n",
"# For validation purposes\n",
"for txt in glob.glob(positive_review_vals):\n",
" with (open(txt, 'r')) as f:\n",
" word_array = raw_to_array(f.read())\n",
" highest_review_word_count = max(highest_review_word_count, len(word_array))\n",
" positive_testrev.append(word_array)\n",
"\n",
"for txt in glob.glob(negative_review_vals):\n",
" with (open(txt, 'r')) as f:\n",
" word_array = raw_to_array(f.read())\n",
" highest_review_word_count = max(highest_review_word_count, len(word_array))\n",
" negative_testrev.append(raw_to_array(f.read()))\n",
" \n",
"\n",
"print('highest word count: ', highest_review_word_count)\n",
"print(positive_reviews[0][:20])\n",
"\n",
"# Plotting length\n",
"training_review_length = np.sort(np.array(training_review_length))\n",
"print('mean of review length: ', np.mean(training_review_length))\n",
"plt.plot(training_review_length)"
]
},
{
"cell_type": "markdown",
"metadata": {
"nbpresent": {
"id": "1fb917cc-0afc-424f-8b88-400a90ad6819"
}
},
"source": [
"## Step 2: Assign vector to vocabs"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "43426664-5c77-45e8-8e50-bd112537d092"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"('Vocab size: ', 75077)\n"
]
}
],
"source": [
"from collections import defaultdict\n",
"\n",
"vocab = defaultdict(float)\n",
"for review in positive_reviews + negative_reviews + positive_testrev + negative_testrev:\n",
" for word in review:\n",
" vocab[word] += 1\n",
" \n",
"print(\"Vocab size: \", len(vocab))"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "088e0cfc-b19d-480a-87e1-73539d08efcd"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"(['vocab_size', 116046], ['layer1_size', 100])\n",
"('vocab hit: ', 58744)\n"
]
}
],
"source": [
"import sys\n",
"word_vecs = {}\n",
"vocab_hit = 0\n",
"\n",
"# Code copies from yoonkim CNN's process_data.py\n",
"with open(word_vector_bin_file, 'rb') as f:\n",
" header = f.readline()\n",
" vocab_size, layer1_size = map(int, header.split())\n",
" print(['vocab_size', vocab_size], ['layer1_size', layer1_size])\n",
"# word_vector_dims = layer1_size\n",
" binary_len = np.dtype('float32').itemsize * layer1_size\n",
" for line in range(vocab_size):\n",
" word = []\n",
" while True:\n",
" ch = f.read(1)\n",
" if ch == ' ':\n",
" word = ''.join(word).lower()\n",
" break\n",
" if ch != '\\n':\n",
" word.append(ch)\n",
" if word in vocab:\n",
" word_vecs[word] = np.fromstring(f.read(binary_len), dtype='float32')\n",
" vocab_hit += 1\n",
" else:\n",
" f.read(binary_len)\n",
"\n",
"print('vocab hit: ', vocab_hit)"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "170beea1-a455-4375-b6fb-4c84778b2a4a"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[ 0.20345445 -0.3446292 -0.79000551 0.50081944 0.44940072 -0.06276506\n",
" -0.22317918 -0.25703317 -0.12849531 -0.31195965 -0.5679732 0.15336281\n",
" -0.27494311 0.20641154 -0.37620348 -0.18761899 -0.02728397 0.1838602\n",
" 0.47868598 -0.49992049 -0.15919879 0.36610898 -0.26038209 0.11431786\n",
" -0.00538804 0.33008578 -0.26670018 0.08674013 -0.55320936 -0.13692944\n",
" 0.21666591 0.1827794 0.26853284 0.17944637 0.63986886 -0.23207539\n",
" 0.13950627 -0.25690758 0.8048203 -0.22281611 -0.30845243 0.40280464\n",
" -0.09700917 -0.10624079 -0.63829452 0.06724913 0.58268034 0.71300983\n",
" 0.21445699 0.07173178 -0.10021135 0.052795 0.21011689 0.55990034\n",
" 0.3317692 -0.1204156 0.33604714 -0.1214296 0.06541372 0.11289533\n",
" -0.33093852 -0.36012915 0.07092347 -0.05363626 0.05315193 -0.40072322\n",
" -0.37435323 0.54634482 0.04995376 0.4135814 0.11987102 0.35045838\n",
" -0.22226219 -0.00525038 0.23082729 0.25800541 0.19155909 -0.05843235\n",
" -0.24793234 0.16106944 0.21992229 -0.36523741 -0.85971153 -0.15568981\n",
" -0.27584159 -0.13528582 0.07193217 -0.0341942 -0.03699406 -0.10003386\n",
" 0.16112548 -0.27273464 0.14367266 -0.05410241 0.15156566 -0.09456853\n",
" 0.33454007 0.14035556 0.5199551 0.19529557]\n"
]
}
],
"source": [
"print(word_vecs['awesome'])"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "cdb3fecf-044e-49af-a897-5a213e02698c"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"('missing vocabs: ', [u'tsukino', u'natilie', u'nordisk', u'happeningb', u'hmoney', u'sangreal', u'bedknob', u'gabreil', u'remendados', u'umte', u'bocka', u'enchelada', u'spracht', u'rumbustious', u'familiarness', u'bendar', u'macmurphy', u'donger', u'bendan', u'vibrational'])\n"
]
}
],
"source": [
"# Check if there are words missing from vector \n",
"# (may happen if we use other word vectors than the one we trained)\n",
"missing_vocabs = []\n",
"for word in vocab:\n",
" if word not in word_vecs and vocab[word] >= 1:\n",
" word_vecs[word] = np.random.uniform(-0.25, 0.25, word_vector_dims)\n",
" missing_vocabs.append(word)\n",
"\n",
"print('missing vocabs: ', missing_vocabs[:20])"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "3817220e-8d9c-49d3-9d64-a48c03ea2430"
}
},
"outputs": [],
"source": [
"total_reviews = len(positive_reviews) + len(negative_reviews)\n",
"max_test = 12500\n",
"\n",
"X_train = []\n",
"Y_train = []\n",
"X_test = []\n",
"Y_test = []\n",
"\n",
"# Due to memory constraint, we only take vectors for first 200 words in movie reviews\n",
"for i, review in enumerate(positive_reviews):\n",
"# word_vector_array = np.ndarray(shape=(max_sentence_length, word_vector_dims), dtype='float32')\n",
" word_vector_array = np.full((max_sentence_length, word_vector_dims), 0, dtype='float32')\n",
" for j, word in enumerate(review):\n",
" word_vector_array[j] = word_vecs[word][:word_vector_dims]\n",
" if j == max_sentence_length -1:\n",
" break\n",
" X_train.append(word_vector_array)\n",
" Y_train.append([0, 1])\n",
" \n",
"for i, review in enumerate(negative_reviews):\n",
" word_vector_array = np.full((max_sentence_length, word_vector_dims), 0, dtype='float32')\n",
" for j, word in enumerate(review):\n",
" word_vector_array[j] = word_vecs[word][:word_vector_dims]\n",
" if j == max_sentence_length - 1:\n",
" break\n",
" X_train.append(word_vector_array)\n",
" Y_train.append([1, 0])\n",
" \n",
"# For validation purposes\n",
"for i, review in enumerate(positive_testrev):\n",
" word_vector_array = np.full((max_sentence_length, word_vector_dims), 0, dtype='float32')\n",
" for j, word in enumerate(review):\n",
" word_vector_array[j] = word_vecs[word][:word_vector_dims]\n",
" if j == max_sentence_length -1:\n",
" break\n",
" X_test.append(word_vector_array)\n",
" Y_test.append([0, 1])\n",
" if i == max_test:\n",
" break\n",
" \n",
"for i, review in enumerate(negative_testrev):\n",
" word_vector_array = np.full((max_sentence_length, word_vector_dims), 0, dtype='float32')\n",
" for j, word in enumerate(review):\n",
" word_vector_array[j] = word_vecs[word][:word_vector_dims]\n",
" if j == max_sentence_length - 1:\n",
" break\n",
" X_test.append(word_vector_array)\n",
" Y_test.append([1, 0])\n",
" if i == max_test:\n",
" break\n",
"\n",
"# print(X_train[0][0][100])\n",
"# print(Y_train[0].shape)\n",
"\n",
"X_train = np.array(X_train)\n",
"Y_train = np.array(Y_train)\n",
"X_test = np.array(X_test)\n",
"Y_test = np.array(Y_test)"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "5941a2a4-8f01-4077-919e-2f1d724e3cff"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"(25000, 200, 100)\n",
"(25000, 2)\n"
]
}
],
"source": [
"print(X_train.shape)\n",
"print(Y_train.shape)"
]
},
{
"cell_type": "markdown",
"metadata": {
"nbpresent": {
"id": "70003047-3373-4d72-87f2-5708a31122c3"
}
},
"source": [
"## Step 3: Keras\n",
"\n",
"![YoonKim CNN Architecture](yoonkim-cnn-architecture.png)"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "3ba15d56-a8a4-4d7a-8c7f-9465f81ac6a4"
}
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Using Theano backend.\n"
]
}
],
"source": [
"from keras.models import Model, Sequential\n",
"from keras.layers.convolutional import Convolution1D, Convolution2D\n",
"from keras.layers.pooling import MaxPooling1D, MaxPooling2D\n",
"from keras.layers import Merge, Dense, Dropout, Activation, Input, Flatten\n",
"from keras.optimizers import SGD"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "7bf80616-2457-4d9d-b3c9-dc7050365dfa"
},
"scrolled": false
},
"outputs": [],
"source": [
"# Based on the paper, there are filters of various sizes\n",
"filters = 150\n",
"epochs = 5\n",
"\n",
"layer1_filter_sizes = [3,4,5]\n",
"layer1_convs = []\n",
"\n",
"graph_in = Input(shape=(max_sentence_length, word_vector_dims))\n",
"\n",
"for filter_size in layer1_filter_sizes:\n",
" conv = Convolution1D(filters,\n",
" filter_size,\n",
" border_mode = 'valid',\n",
" activation='relu',\n",
" subsample_length=1)(graph_in)\n",
" pool = MaxPooling1D(pool_length=2)(conv)\n",
" flatten = Flatten()(pool)\n",
" layer1_convs.append(flatten)\n",
"\n",
"# Merge the conv\n",
"merged = Merge(mode='concat')(layer1_convs)\n",
"graph = Model(input=graph_in, output=merged)\n",
"\n",
"final_model = Sequential()\n",
"final_model.add(graph)\n",
"# final_model.add(Dense(32))\n",
"# final_model.add(Activation('relu'))\n",
"# final_model.add(Dropout(0.25))\n",
"final_model.add(Dense(16))\n",
"final_model.add(Activation('relu'))\n",
"final_model.add(Dropout(0.5))\n",
"final_model.add(Dense(2))\n",
"final_model.add(Activation('softmax'))\n",
"\n",
"sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)\n",
"final_model.compile(loss='binary_crossentropy',\n",
" optimizer='rmsprop',\n",
" metrics=['accuracy'])\n"
]
},
{
"cell_type": "markdown",
"metadata": {
"nbpresent": {
"id": "5a605fa8-fc27-44b6-b98f-51ab208c9930"
}
},
"source": [
"## Step 4: TRAIN THIS"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {
"collapsed": false,
"nbpresent": {
"id": "082122aa-9d01-40d0-bac7-8476a8e2098f"
}
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Train on 25000 samples, validate on 25000 samples\n",
"Epoch 1/5\n",
"25000/25000 [==============================] - 22s - loss: 0.0756 - acc: 0.9823 - val_loss: 1.1921e-07 - val_acc: 1.0000\n",
"Epoch 2/5\n",
"25000/25000 [==============================] - 22s - loss: 0.0534 - acc: 0.9834 - val_loss: 1.1921e-07 - val_acc: 1.0000\n",
"Epoch 3/5\n",
"25000/25000 [==============================] - 22s - loss: 0.0528 - acc: 0.9827 - val_loss: 1.1921e-07 - val_acc: 1.0000\n",
"Epoch 4/5\n",
"25000/25000 [==============================] - 23s - loss: 0.0495 - acc: 0.9842 - val_loss: 1.1921e-07 - val_acc: 1.0000\n",
"Epoch 5/5\n",
"25000/25000 [==============================] - 22s - loss: 0.0453 - acc: 0.9861 - val_loss: 1.1921e-07 - val_acc: 1.0000\n"
]
},
{
"data": {
"text/plain": [
"<keras.callbacks.History at 0x7fc2cad44e10>"
]
},
"execution_count": 12,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"final_model.fit(X_train, \n",
" Y_train, \n",
" batch_size=32,\n",
" nb_epoch=epochs,\n",
" validation_data=[X_test, Y_test])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"**!!** Something is wrong. The validation accuracy is too high."
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {
"collapsed": false
},
"outputs": [],
"source": [
"final_model.save('jarvis.h5')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Step 5: Custom Predict\n",
"\n",
"Will implement for small bit of text prediction later.\n",
"\n",
"Somehow the predictions are dead wrong here."
]
},
{
"cell_type": "code",
"execution_count": 46,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"def test_predict(text):\n",
" word_array = raw_to_array(text)\n",
" word_vec_array = np.full(fill_value=0.0,\n",
" shape=(1, max_sentence_length, word_vector_dims),\n",
" dtype='float32')\n",
" for i, word in enumerate(word_array):\n",
" if word in word_vecs:\n",
" word_vec_array[0][i] = np.array(word_vecs[word][:word_vector_dims])\n",
" else:\n",
" word_vec_array[0][i] = np.random.uniform(-0.25, 0.25, word_vector_dims) \n",
" if i + 1 >= max_sentence_length:\n",
" break\n",
"# return final_model.predict(word_vec_array, batch_size=1)\n",
" prediction = final_model.predict(word_vec_array, batch_size=1)[0]\n",
" if prediction[1] > prediction[0]:\n",
" return ['movie review is positive', prediction]\n",
" else:\n",
" return ['movie review is negative', prediction]"
]
},
{
"cell_type": "code",
"execution_count": 63,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is positive',\n",
" array([ 5.67092002e-06, 9.99994278e-01], dtype=float32)]"
]
},
"execution_count": 63,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict(\"OMG this warcraft film adaptation was so awesome\")"
]
},
{
"cell_type": "code",
"execution_count": 64,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is negative', array([ 0.95606565, 0.04393429], dtype=float32)]"
]
},
"execution_count": 64,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict(\"gosh this is just bad\")"
]
},
{
"cell_type": "code",
"execution_count": 65,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is positive',\n",
" array([ 1.83029895e-14, 1.00000000e+00], dtype=float32)]"
]
},
"execution_count": 65,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict(\"train to busan was one of the most value for money movie one will ever pay for\")"
]
},
{
"cell_type": "code",
"execution_count": 50,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is positive',\n",
" array([ 4.78235685e-04, 9.99521732e-01], dtype=float32)]"
]
},
"execution_count": 50,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict(\"Blackhat is not only disappointing, its embarrassing\")"
]
},
{
"cell_type": "code",
"execution_count": 51,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is positive', array([ 0., 1.], dtype=float32)]"
]
},
"execution_count": 51,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict('''\n",
"Suffers from inconsistencies, both technical and story wise. They change the shooting styles, cameras, fps, warmth/cold - for no apparent reasons at all. Feels like it's not clear what this movie \"wants to be\". The main character is supposed to be a \"super-hacker\" but doesn't do anything \"super hack-y\", just wanders around, shooting people, and nails the female protagonist. Doesn't have many hacking-scenes for a \"hackers movie\", has tons of boring gun-scenes instead, from some reason. The motivation of the villain was, not interesting. References many other \"movie-cliches\" (not in a good way). Severely lacks humor. The few jokes in it are really cheesy (yeah, it's not a comedy , but comic reliefs are important). Many of the audience members left the theater in the middle or before the end\n",
"''')"
]
},
{
"cell_type": "code",
"execution_count": 52,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is positive',\n",
" array([ 2.26467456e-10, 1.00000000e+00], dtype=float32)]"
]
},
"execution_count": 52,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict('explosive summer flick that will keep you on the couch for hours')"
]
},
{
"cell_type": "code",
"execution_count": 53,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is negative', array([ 0.61056298, 0.38943696], dtype=float32)]"
]
},
"execution_count": 53,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict('why would anyone watch this?')"
]
},
{
"cell_type": "code",
"execution_count": 54,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is positive', array([ 0.0224589 , 0.97754115], dtype=float32)]"
]
},
"execution_count": 54,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict('some people actually walked out from cinema')"
]
},
{
"cell_type": "code",
"execution_count": 55,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is positive',\n",
" array([ 7.43133866e-09, 1.00000000e+00], dtype=float32)]"
]
},
"execution_count": 55,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict('this is definitely the best flick from christopher nolan yet!')"
]
},
{
"cell_type": "code",
"execution_count": 56,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is negative', array([ 0.9827314 , 0.01726856], dtype=float32)]"
]
},
"execution_count": 56,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict('i dug my eyes out')"
]
},
{
"cell_type": "code",
"execution_count": 57,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is positive',\n",
" array([ 1.04985540e-08, 1.00000000e+00], dtype=float32)]"
]
},
"execution_count": 57,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict('seriously this is the one you must watch this year')"
]
},
{
"cell_type": "code",
"execution_count": 62,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is negative',\n",
" array([ 9.99516845e-01, 4.83212119e-04], dtype=float32)]"
]
},
"execution_count": 62,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict('touching love story indeed')"
]
},
{
"cell_type": "code",
"execution_count": 67,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is positive',\n",
" array([ 3.33922645e-09, 1.00000000e+00], dtype=float32)]"
]
},
"execution_count": 67,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict('would love to lie on the grassfield and watch this with her again')"
]
},
{
"cell_type": "code",
"execution_count": 69,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"['movie review is positive',\n",
" array([ 2.70553046e-12, 1.00000000e+00], dtype=float32)]"
]
},
"execution_count": 69,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"test_predict('I bet there are more productive things to do than watching this film')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": []
}
],
"metadata": {
"anaconda-cloud": {},
"kernelspec": {
"display_name": "Python [tensorflow]",
"language": "python",
"name": "Python [tensorflow]"
},
"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"
},
"nbpresent": {
"slides": {
"37e74d51-8e3e-45ef-a6e1-8a70764d8838": {
"id": "37e74d51-8e3e-45ef-a6e1-8a70764d8838",
"prev": "ba25d42f-178f-4bff-878f-237edd09c5f5",
"regions": {
"02d29844-ddc7-4301-9e0f-8cf2188ea9a7": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "cdb3fecf-044e-49af-a897-5a213e02698c",
"part": "whole"
},
"id": "02d29844-ddc7-4301-9e0f-8cf2188ea9a7"
}
},
"theme": null
},
"3961229a-9df7-4080-8adb-51386f2bc3bd": {
"id": "3961229a-9df7-4080-8adb-51386f2bc3bd",
"prev": "53407f4b-429c-4df1-9d5d-c26d568e5a4f",
"regions": {
"62e52504-c547-4256-809d-50995400803a": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "789db32e-3bb2-49b4-8ecc-e8f9f0d67a7b",
"part": "whole"
},
"id": "62e52504-c547-4256-809d-50995400803a"
}
}
},
"53407f4b-429c-4df1-9d5d-c26d568e5a4f": {
"id": "53407f4b-429c-4df1-9d5d-c26d568e5a4f",
"prev": "af0c81d8-7d10-422d-a598-1cb99dee2e2c",
"regions": {
"3222cf87-ef46-4651-8cf2-80a78e9eca3b": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "bc104c55-9cb1-4d4c-a1bf-363277a69da8",
"part": "whole"
},
"id": "3222cf87-ef46-4651-8cf2-80a78e9eca3b"
}
}
},
"64728ebe-592c-4d5b-baa4-dc089e2704fb": {
"id": "64728ebe-592c-4d5b-baa4-dc089e2704fb",
"prev": "99bda4b6-a416-4ffc-9fa9-77a771dd27ec",
"regions": {
"8f248628-70ae-48f5-9bd1-eb1cfdf6e2f6": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "07039836-86fc-4567-b302-ef9265af8307",
"part": "whole"
},
"id": "8f248628-70ae-48f5-9bd1-eb1cfdf6e2f6"
}
}
},
"875c94af-0074-4c3e-8c45-e90c2ed94612": {
"id": "875c94af-0074-4c3e-8c45-e90c2ed94612",
"prev": "fb703c45-7b3e-47cb-a3cb-90072b339ac7",
"regions": {
"1dd617cb-85a2-4e96-8f9d-a180b24413b1": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "088e0cfc-b19d-480a-87e1-73539d08efcd",
"part": "whole"
},
"id": "1dd617cb-85a2-4e96-8f9d-a180b24413b1"
}
}
},
"8ce9e396-9e36-48d8-8dad-652e0874f01c": {
"id": "8ce9e396-9e36-48d8-8dad-652e0874f01c",
"prev": "eeb7b456-444b-44b5-a9ce-47c172546952",
"regions": {
"851f0ea5-8a8c-4a2b-9105-1d56ba2ef2d7": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "082122aa-9d01-40d0-bac7-8476a8e2098f",
"part": "whole"
},
"id": "851f0ea5-8a8c-4a2b-9105-1d56ba2ef2d7"
}
}
},
"905771fb-bf2d-42a7-aec2-5da33f44c325": {
"id": "905771fb-bf2d-42a7-aec2-5da33f44c325",
"prev": "8ce9e396-9e36-48d8-8dad-652e0874f01c",
"regions": {
"63519927-4a11-499b-b1a4-ac6dbd9e8bb0": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "3a8186e0-2504-4f64-9086-fe7ab528f2e3",
"part": "whole"
},
"id": "63519927-4a11-499b-b1a4-ac6dbd9e8bb0"
}
}
},
"91908d9f-ed0e-4588-a13b-308c73734071": {
"id": "91908d9f-ed0e-4588-a13b-308c73734071",
"prev": "905771fb-bf2d-42a7-aec2-5da33f44c325",
"regions": {
"52d4555d-d45e-4115-98e4-21d8340c2257": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "60e00169-ee44-4c0c-9c55-7121d4c8332f",
"part": "whole"
},
"id": "52d4555d-d45e-4115-98e4-21d8340c2257"
}
}
},
"99bda4b6-a416-4ffc-9fa9-77a771dd27ec": {
"id": "99bda4b6-a416-4ffc-9fa9-77a771dd27ec",
"prev": null,
"regions": {
"3e966305-d8cd-4860-8194-9e4460aebbfe": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.0833250996187788,
"y": 0.10830039485643014
},
"id": "3e966305-d8cd-4860-8194-9e4460aebbfe"
},
"81e1e2bd-e729-4e37-8913-bf2162a507bd": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.03863636659710592,
"y": 0.09407114653112136
},
"content": {
"cell": "aa8439c7-cf04-49a6-84f3-2e8af08aca7f",
"part": "whole"
},
"id": "81e1e2bd-e729-4e37-8913-bf2162a507bd"
}
}
},
"9aeefc9a-ead7-4fff-b68f-12374636e8c3": {
"id": "9aeefc9a-ead7-4fff-b68f-12374636e8c3",
"prev": "dc88abf0-a5b8-43d9-b06f-0f6172443154",
"regions": {
"11975b9d-2c03-4d62-b3f8-273c4a929ac5": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "5941a2a4-8f01-4077-919e-2f1d724e3cff",
"part": "whole"
},
"id": "11975b9d-2c03-4d62-b3f8-273c4a929ac5"
}
}
},
"a4bcb0ed-33c6-45a4-a6b6-6fda4109f64f": {
"id": "a4bcb0ed-33c6-45a4-a6b6-6fda4109f64f",
"prev": "3961229a-9df7-4080-8adb-51386f2bc3bd",
"regions": {
"2cbebc16-1a38-4f7c-ad9a-ad220db57799": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "1fb917cc-0afc-424f-8b88-400a90ad6819",
"part": "whole"
},
"id": "2cbebc16-1a38-4f7c-ad9a-ad220db57799"
}
}
},
"af0c81d8-7d10-422d-a598-1cb99dee2e2c": {
"id": "af0c81d8-7d10-422d-a598-1cb99dee2e2c",
"prev": "64728ebe-592c-4d5b-baa4-dc089e2704fb",
"regions": {
"934404bd-cff4-4c2d-a5c2-093f7660a160": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "35ac46e5-404e-4d92-b9bd-073c346223a0",
"part": "whole"
},
"id": "934404bd-cff4-4c2d-a5c2-093f7660a160"
}
}
},
"ba25d42f-178f-4bff-878f-237edd09c5f5": {
"id": "ba25d42f-178f-4bff-878f-237edd09c5f5",
"prev": "875c94af-0074-4c3e-8c45-e90c2ed94612",
"regions": {
"39ec72be-7f2a-49bc-a862-d0ddaf0741fa": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "170beea1-a455-4375-b6fb-4c84778b2a4a",
"part": "whole"
},
"id": "39ec72be-7f2a-49bc-a862-d0ddaf0741fa"
}
}
},
"bd7a242b-b4ee-4781-abd6-45dc7d6604fb": {
"id": "bd7a242b-b4ee-4781-abd6-45dc7d6604fb",
"prev": "91908d9f-ed0e-4588-a13b-308c73734071",
"regions": {
"dd046f4e-2e3d-444a-9d20-5929545381d9": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "a1b9ac65-1c0d-436c-8c4c-6aec5c59e4e9",
"part": "whole"
},
"id": "dd046f4e-2e3d-444a-9d20-5929545381d9"
}
}
},
"c08bcf99-f04e-494d-bc58-e666d4ec92ba": {
"id": "c08bcf99-f04e-494d-bc58-e666d4ec92ba",
"prev": "d5f1666e-7e38-4606-8790-1cd4b70da45e",
"regions": {
"b561ca37-037f-454c-a483-740986aa86d9": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "3ba15d56-a8a4-4d7a-8c7f-9465f81ac6a4",
"part": "whole"
},
"id": "b561ca37-037f-454c-a483-740986aa86d9"
}
}
},
"cff711ff-48f0-4015-b6af-b53eef839e16": {
"id": "cff711ff-48f0-4015-b6af-b53eef839e16",
"prev": "bd7a242b-b4ee-4781-abd6-45dc7d6604fb",
"regions": {
"e90e7eda-38c0-49b8-b941-042573be21a2": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "041f0afc-f01c-42ae-8937-a81d76ccc74b",
"part": "whole"
},
"id": "e90e7eda-38c0-49b8-b941-042573be21a2"
}
}
},
"d393470b-f821-447b-b90a-8a873f5615c6": {
"id": "d393470b-f821-447b-b90a-8a873f5615c6",
"prev": "c08bcf99-f04e-494d-bc58-e666d4ec92ba",
"regions": {
"010c3eef-e5a8-4833-b61d-5d9497a35284": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "7bf80616-2457-4d9d-b3c9-dc7050365dfa",
"part": "whole"
},
"id": "010c3eef-e5a8-4833-b61d-5d9497a35284"
}
}
},
"d5f1666e-7e38-4606-8790-1cd4b70da45e": {
"id": "d5f1666e-7e38-4606-8790-1cd4b70da45e",
"prev": "9aeefc9a-ead7-4fff-b68f-12374636e8c3",
"regions": {
"5683aaa1-f7d4-49a9-a332-b8e6b72e355c": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "70003047-3373-4d72-87f2-5708a31122c3",
"part": "whole"
},
"id": "5683aaa1-f7d4-49a9-a332-b8e6b72e355c"
}
}
},
"dc88abf0-a5b8-43d9-b06f-0f6172443154": {
"id": "dc88abf0-a5b8-43d9-b06f-0f6172443154",
"prev": "37e74d51-8e3e-45ef-a6e1-8a70764d8838",
"regions": {
"e0e421c5-9d16-4e11-991f-3010779f786f": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "3817220e-8d9c-49d3-9d64-a48c03ea2430",
"part": "whole"
},
"id": "e0e421c5-9d16-4e11-991f-3010779f786f"
}
}
},
"eeb7b456-444b-44b5-a9ce-47c172546952": {
"id": "eeb7b456-444b-44b5-a9ce-47c172546952",
"prev": "d393470b-f821-447b-b90a-8a873f5615c6",
"regions": {
"c89b5a8c-d9fc-4756-b536-0340b41e8d41": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "5a605fa8-fc27-44b6-b98f-51ab208c9930",
"part": "whole"
},
"id": "c89b5a8c-d9fc-4756-b536-0340b41e8d41"
}
}
},
"fb703c45-7b3e-47cb-a3cb-90072b339ac7": {
"id": "fb703c45-7b3e-47cb-a3cb-90072b339ac7",
"prev": "a4bcb0ed-33c6-45a4-a6b6-6fda4109f64f",
"regions": {
"fe6d4f25-6a25-4a89-981d-509609414a42": {
"attrs": {
"height": 0.8,
"width": 0.8,
"x": 0.1,
"y": 0.1
},
"content": {
"cell": "43426664-5c77-45e8-8e50-bd112537d092",
"part": "whole"
},
"id": "fe6d4f25-6a25-4a89-981d-509609414a42"
}
}
}
},
"themes": {}
}
},
"nbformat": 4,
"nbformat_minor": 1
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment