def train(text, epochs=100, save_freq=10):
# character to index and index to char mappings
char_to_idx = {ch: i for (i, ch) in enumerate(sorted(list(set(text))))}
print("Number of unique characters: " + str(len(char_to_idx))) #86
with open(os.path.join(DATA_DIR, 'char_to_idx.json'), 'w') as f:
json.dump(char_to_idx, f)
idx_to_char = {i: ch for (ch, i) in char_to_idx.items()}
vocab_size = len(char_to_idx)
#model_architecture
model = build_model(BATCH_SIZE, SEQ_LENGTH, vocab_size)
model.summary()
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
#Train data generation
T = np.asarray(
[char_to_idx[c] for c in text],
dtype=np.int32) #convert complete text into numerical indices
print("Length of text:" + str(T.size)) #129,665
steps_per_epoch = (len(text) / BATCH_SIZE - 1) / SEQ_LENGTH
log = TrainLogger('training_log.csv')
for epoch in range(epochs):
print('\nEpoch {}/{}'.format(epoch + 1, epochs))
losses, accs = [], []
for i, (X, Y) in enumerate(read_batches(T, vocab_size)):
print(X)
loss, acc = model.train_on_batch(X, Y)
print('Batch {}: loss = {}, acc = {}'.format(i + 1, loss, acc))
losses.append(loss)
accs.append(acc)
log.add_entry(np.average(losses), np.average(accs))
if (epoch + 1) % save_freq == 0:
save_weights(epoch + 1, model)
print('Saved checkpoint to', 'weights.{}.h5'.format(epoch + 1))
Bidirectional(GRU(units=50, return_sequences=True)),
tfa.layers.GroupNormalization(),
Dropout(0.2),
Bidirectional(GRU(units=50, return_sequences=True)),
tfa.layers.GroupNormalization(),
Dropout(0.2),
Bidirectional(LSTM(units=50)),
tfa.layers.GroupNormalization(),
Dropout(0.2),
Dense(units=1, activation='sigmoid')
]
name = 'bgru_lstm'
epochs = 100
batch_size = 32
# Training the model.
model = rnn_model.build_model(input_shape, layers)
model, history = rnn_model.train(model,
name,
x_train,
y_train,
epochs=epochs,
batch_size=batch_size)
# Making predictions.
y_predict = model.predict(x_test)
# Visualising the prediction.
stocks.plot_prediction(y_test, y_predict, 'Real Tesla Stock Price',
'Predicted Tesla Stock Price')