def test_SimpleSeq2Seq():
x = np.random.random((samples, input_length, input_dim))
y = np.random.random((samples, output_length, output_dim))
models = []
models += [
SimpleSeq2Seq(output_dim=output_dim,
output_length=output_length,
input_shape=(input_length, input_dim))
]
models += [
SimpleSeq2Seq(output_dim=output_dim,
output_length=output_length,
input_shape=(input_length, input_dim),
depth=2)
]
for model in models:
model.compile(loss='mse', optimizer='sgd')
model.fit(x, y, nb_epoch=1)
def test_SimpleSeq2Seq():
x = np.random.random((samples, input_length, input_dim))
y = np.random.random((samples, output_length, output_dim))
models = []
models += [SimpleSeq2Seq(output_dim=output_dim, hidden_dim=hidden_dim, output_length=output_length, input_shape=(input_length, input_dim))]
#models += [SimpleSeq2Seq(output_dim=output_dim, hidden_dim=hidden_dim, output_length=output_length, input_shape=(input_length, input_dim), depth=2)]
for model in models:
model.compile(loss='mse', optimizer='sgd')
plot_model(model, to_file='model3.png', show_shapes=True)
def test_SimpleSeq2Seq():
x = np.random.random((batch, max_encoder_length, input_dim))
y = np.random.random((batch, max_decoder_length, output_dim))
models = []
models += [
SimpleSeq2Seq(output_dim=output_dim,
hidden_dim=hidden_dim,
output_length=max_decoder_length,
input_shape=(max_encoder_length, input_dim))
]
models += [
SimpleSeq2Seq(output_dim=output_dim,
hidden_dim=hidden_dim,
output_length=max_decoder_length,
input_shape=(max_encoder_length, input_dim),
depth=2)
]
for model in models:
model.compile(loss='mse', optimizer='sgd')
model.fit(x, y, nb_epoch=1)
# turn x and y into numpy array among other things
train_x = pad_sequences(train_x, maxlen=maxlen_x)
train_y = pad_sequences(train_y, maxlen=maxlen_y)
print train_y.shape
print train_y
# convert train_y into (num_examples, maxlen, alphabet_size)
# train_y = to_categorical(np.array(train_y), classes)
model = Sequential()
model.add(Embedding(input_dim=len(dataset.input2int),
output_dim=cfg.getint('cnn', 'embdims'),
input_length=maxlen_x))
#model.add(SimpleSeq2Seq(input_dim=5, hidden_dim=10,
# output_length=8, output_dim=8))
model.add(SimpleSeq2Seq(hidden_dim=10, output_length=maxlen_y, output_dim=len(dataset.output2int)))
model.compile(loss='mse', optimizer='rmsprop')
model.fit(train_x,
train_y,
epochs=cfg.getint('cnn', 'epochs'),
batch_size=cfg.getint('cnn', 'batch'),
verbose=1,
validation_split=0.0)
# get_weights() returns a list of 1 element
# dump these weights to file (word2vec model format)
weights = model.layers[0].get_weights()[0]
word2vec.write_vectors(dataset.word2int, weights, 'weights.txt')
# probability for each class; (test size, num of classes)
partial_train_data = np.concatenate([
source_data[:i * num_val_samples],
source_data[(i + 1) * num_val_samples:]
],
axis=0)
partial_train_targets = np.concatenate([
target_data[:i * num_val_samples],
target_data[(i + 1) * num_val_samples:]
],
axis=0)
# Build the Keras model (already compiled)
model = SimpleSeq2Seq(input_dim=9,
input_length=3,
hidden_dim=10,
output_length=1,
output_dim=1,
dropout=0.3,
depth=3)
# model = Seq2Seq(input_dim=9, input_length=3,hidden_dim=10, output_length=1, output_dim=1, depth=4)
# model = AttentionSeq2Seq(input_dim=9, input_length=3, hidden_dim=10, output_length=1, output_dim=1, depth=3)
model.compile(loss='mse', optimizer='adam')
# Train the model (in silent mode, verbose=0)
model.fit(partial_train_data,
partial_train_targets,
epochs=num_epochs,
batch_size=128,
verbose=1)
# Evaluate the model on the validation data