Skip to content

Commit

Permalink
LSTM
Browse files Browse the repository at this point in the history
  • Loading branch information
@nursnaaz
nursnaaz committed Sep 2, 2019
1 parent 05235fc commit 2a5914f
Show file tree
Hide file tree
Showing 24 changed files with 8,836 additions and 0 deletions.
Binary file added 02 - Day - 2 Convolution NN/4. YOLO-OpenCV/yolov3-coco/.DS_Store
View file
Binary file not shown.
550 changes: 550 additions & 0 deletions 03 - Day - 3 Reccurent NN/1. Basic/01. RNN Structure.ipynb
View file

Large diffs are not rendered by default.

594 changes: 594 additions & 0 deletions 03 - Day - 3 Reccurent NN/1. Basic/02. Simple RNN.ipynb
View file

Large diffs are not rendered by default.

357 changes: 357 additions & 0 deletions 03 - Day - 3 Reccurent NN/2. Advanced/1. Bi-Directional LSTM.ipynb
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,357 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Using TensorFlow backend.\n"
]
}
],
"source": [
"import numpy as np\n",
"\n",
"from sklearn.metrics import accuracy_score\n",
"from keras.datasets import reuters\n",
"from keras.preprocessing.sequence import pad_sequences\n",
"from keras.utils import to_categorical"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"# parameters for data load\n",
"num_words = 30000\n",
"maxlen = 50\n",
"test_split = 0.3"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"(X_train, y_train), (X_test, y_test) = reuters.load_data(num_words = num_words, maxlen = maxlen, test_split = test_split)"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"# pad the sequences with zeros \n",
"# padding parameter is set to 'post' => 0's are appended to end of sequences\n",
"X_train = pad_sequences(X_train, padding = 'post')\n",
"X_test = pad_sequences(X_test, padding = 'post')\n",
"\n",
"X_train = np.array(X_train).reshape((X_train.shape[0], X_train.shape[1], 1))\n",
"X_test = np.array(X_test).reshape((X_test.shape[0], X_test.shape[1], 1))\n",
"\n",
"y_data = np.concatenate((y_train, y_test))\n",
"y_data = to_categorical(y_data)\n",
"y_train = y_data[:1395]\n",
"y_test = y_data[1395:]"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 1. Deep RNN\n",
"- RNNs can be made deep, with multiple layers, like CNNs or MLPs\n",
"- Beware that RNNs take long to train compared to CNNs"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"from keras.models import Sequential\n",
"from keras.layers import Dense, LSTM, Activation\n",
"from keras import optimizers\n",
"from keras.wrappers.scikit_learn import KerasClassifier"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"def deep_lstm():\n",
" model = Sequential()\n",
" model.add(LSTM(20, input_shape = (49,1), return_sequences = True))\n",
" model.add(LSTM(20, return_sequences = True))\n",
" model.add(LSTM(20, return_sequences = True))\n",
" model.add(LSTM(20, return_sequences = False))\n",
" model.add(Dense(46))\n",
" model.add(Activation('softmax'))\n",
" \n",
" adam = optimizers.Adam(lr = 0.001)\n",
" model.compile(loss = 'categorical_crossentropy', optimizer = adam, metrics = ['accuracy'])\n",
" \n",
" return model"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"model = KerasClassifier(build_fn = deep_lstm, epochs = 200, batch_size = 50, verbose = 1)\n",
"model.fit(X_train, y_train)"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"599/599 [==============================] - 0s \n"
]
}
],
"source": [
"y_pred = model.predict(X_test)\n",
"y_test_ = np.argmax(y_test, axis = 1)"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"0.816360601002\n"
]
}
],
"source": [
"print(accuracy_score(y_pred, y_test_))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 2. Bidirectional RNN\n",
"- Bidirectional RNNs consider not only one-way influence of sequence, but also the other way\n",
"- Actually, they can be thought as building two separate RNNs, and merging them\\\n",
"<br>\n",
"<img src=\"http://d3kbpzbmcynnmx.cloudfront.net/wp-content/uploads/2015/09/bidirectional-rnn.png\" style=\"width: 400px\"/>\n",
"</br>"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"from keras.layers import Bidirectional"
]
},
{
"cell_type": "code",
"execution_count": 21,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"def bidirectional_lstm():\n",
" model = Sequential()\n",
" model.add(Bidirectional(LSTM(20, return_sequences = False), input_shape = (49,1)))\n",
" model.add(Dense(46))\n",
" model.add(Activation('softmax'))\n",
" \n",
" adam = optimizers.Adam(lr = 0.001)\n",
" model.compile(loss = 'categorical_crossentropy', optimizer = adam, metrics = ['accuracy'])\n",
" \n",
" return model"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"model = KerasClassifier(build_fn = bidirectional_lstm, epochs = 200, batch_size = 50, verbose = 1)\n",
"model.fit(X_train, y_train)"
]
},
{
"cell_type": "code",
"execution_count": 23,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"500/599 [========================>.....] - ETA: 0s"
]
}
],
"source": [
"y_pred = model.predict(X_test)\n",
"y_test_ = np.argmax(y_test, axis = 1)"
]
},
{
"cell_type": "code",
"execution_count": 24,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"0.841402337229\n"
]
}
],
"source": [
"print(accuracy_score(y_pred, y_test_))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 3. Deep Bidirectional RNN\n",
"- Bidirectional RNNs can be stacked\n",
"\n",
"<img src=\"http://www.wildml.com/wp-content/uploads/2015/09/Screen-Shot-2015-09-16-at-2.21.51-PM-272x300.png\" style=\"width: 300px\"/>"
]
},
{
"cell_type": "code",
"execution_count": 27,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"def deep_bidirectional_lstm():\n",
" model = Sequential()\n",
" model.add(Bidirectional(LSTM(10, return_sequences = True), input_shape = (49,1)))\n",
" model.add(Bidirectional(LSTM(10, return_sequences = True)))\n",
" model.add(Bidirectional(LSTM(10, return_sequences = True)))\n",
" model.add(Bidirectional(LSTM(10, return_sequences = False)))\n",
" model.add(Dense(46))\n",
" model.add(Activation('softmax'))\n",
" \n",
" adam = optimizers.Adam(lr = 0.001)\n",
" model.compile(loss = 'categorical_crossentropy', optimizer = adam, metrics = ['accuracy'])\n",
" \n",
" return model"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"model = KerasClassifier(build_fn = deep_bidirectional_lstm, epochs = 200, batch_size = 50, verbose = 1)\n",
"model.fit(X_train, y_train)"
]
},
{
"cell_type": "code",
"execution_count": 29,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"599/599 [==============================] - 1s \n"
]
}
],
"source": [
"y_pred = model.predict(X_test)\n",
"y_test_ = np.argmax(y_test, axis = 1)"
]
},
{
"cell_type": "code",
"execution_count": 30,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"0.824707846411\n"
]
}
],
"source": [
"print(accuracy_score(y_pred, y_test_))"
]
}
],
"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": "3.6.4"
}
},
"nbformat": 4,
"nbformat_minor": 2
}
Loading

0 comments on commit 2a5914f

Please sign in to comment.