train.py

import numpy as np
import pandas as pd
from gensim.models.word2vec import Word2Vec
from keras.preprocessing.sequence import pad_sequences
from keras.models import Sequential
from keras.layers import Activation, Flatten, Dense, Conv1D, Embedding, Input
from keras import backend as K


def loss(y_true, y_pred):
    return(K.mean(K.abs(y_true-y_pred)))


train_df = pd.read_pickle("./train.pkl").sample(frac=1, random_state=123)
test_df = pd.read_pickle("./test.pkl")
w2v_model = Word2Vec.load("./word2vec.model")
train_df.head()


embedding_matrix = np.zeros(
    (len(w2v_model.wv.vocab.items()) + 1, w2v_model.vector_size))
word2idx = {}
vocab_list = [(word, w2v_model.wv[word])
              for word, _ in w2v_model.wv.vocab.items()]

for i, vocab in enumerate(vocab_list):
    word, vec = vocab
    embedding_matrix[i + 1] = vec
    word2idx[word] = i + 1


def text_to_index(corpus):
    new_corpus = []
    for doc in corpus:
        new_doc = []
        for word in doc:
            try:
                new_doc.append(word2idx[word])
            except:
                new_doc.append(0)
        new_corpus.append(new_doc)
    return np.array(new_corpus)


PADDING_LENGTH = 200
X = text_to_index(train_df.text)
X = pad_sequences(X, maxlen=PADDING_LENGTH)
print("Shape:", X.shape)
print("Sample:", X[0])


embedding_layer = Embedding(input_dim=embedding_matrix.shape[0],
                            output_dim=embedding_matrix.shape[1],
                            weights=[embedding_matrix],
                            trainable=False,
                            input_length=200)


model = Sequential()
model.add(embedding_layer)
model.add(Conv1D(
    filters=(9),
    kernel_size=(256),
    padding='same',
    input_shape=(9, 32, 32),
    activation='relu',
))
model.add(Flatten())
model.add(Dense(2, activation='relu'))
model.compile(optimizer='adam',
              loss=loss,
              metrics=['accuracy'])
model.summary()

result = train_df.drop(['text'], axis=1).as_matrix()

model.fit(x=X, y=result, batch_size=3000, epochs=100, validation_split=0.05)


# X_test = text_to_index(test_df.text)
# X_test = pad_sequences(X_test, maxlen=PADDING_LENGTH)

# predictions = model.predict_classes(X_test, verbose=1).flatten()
# submissions = pd.DataFrame(
#     {"id": test_df['id'], "good": predictions[0], "bad": predictions[1]})
# submissions.to_csv("survived.csv", index=False, header=True)