supervised_word_embeddings.ipynb

Kernel: Python 3

import numpy as np
from tensorflow.keras.preprocessing.text import one_hot
from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
from tensorflow.keras.layers import Flatten
from tensorflow.keras.layers import Embedding

reviews = ['nice food',
        'amazing restaurant',
        'too good',
        'just loved it!',
        'will go again',
        'horrible food',
        'never go there',
        'poor service',
        'poor quality',
        'needs improvement']

sentiment = np.array([1,1,1,1,1,0,0,0,0,0])

one_hot("amazing restaurant",30)

[4, 23]

vocab_size = 30
encoded_reviews = [one_hot(d, vocab_size) for d in reviews]
print(encoded_reviews)

[[13, 21], [4, 23], [14, 17], [8, 15, 16], [22, 15, 29], [8, 21], [26, 15, 24], [16, 4], [16, 12], [4, 29]]

max_length = 4
padded_reviews = pad_sequences(encoded_reviews, maxlen=max_length, padding='post')
print(padded_reviews)

[[13 21  0  0]
 [ 4 23  0  0]
 [14 17  0  0]
 [ 8 15 16  0]
 [22 15 29  0]
 [ 8 21  0  0]
 [26 15 24  0]
 [16  4  0  0]
 [16 12  0  0]
 [ 4 29  0  0]]

embeded_vector_size = 5

model = Sequential()
model.add(Embedding(vocab_size, embeded_vector_size, input_length=max_length,name="embedding"))
model.add(Flatten())
model.add(Dense(1, activation='sigmoid'))

X = padded_reviews
y = sentiment

model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
print(model.summary())

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
embedding (Embedding)        (None, 4, 5)              150       
_________________________________________________________________
flatten_1 (Flatten)          (None, 20)                0         
_________________________________________________________________
dense_1 (Dense)              (None, 1)                 21        
=================================================================
Total params: 171
Trainable params: 171
Non-trainable params: 0
_________________________________________________________________
None

model.fit(X, y, epochs=50, verbose=0)

<tensorflow.python.keras.callbacks.History at 0x1bb8daa5a30>

# evaluate the model
loss, accuracy = model.evaluate(X, y)
accuracy

1/1 [==============================] - 0s 1ms/step - loss: 0.6384 - accuracy: 1.0000

1.0

weights = model.get_layer('embedding').get_weights()[0]
len(weights)

30

weights[13]

array([-0.08330977, -0.06752131, -0.04629624, -0.00765801, -0.02024159],
      dtype=float32)

weights[4]

array([-0.07935128, -0.08574004,  0.06615968, -0.02349528,  0.00917289],
      dtype=float32)

weights[16]

array([ 0.0128377 ,  0.03549778,  0.05134471, -0.07147218,  0.03261041],
      dtype=float32)