import pickle
import tsp_data as tsp
import numpy as np
from tensorflow import keras
import tensorflow as tf
def scheduler(epoch):
if epoch < nb_epochs/4:
return learning_rates
elif epoch < nb_epochs/2:
return learning_rate*0.5
return learning_rate*0.1
if __name__=='__main__':
print("preparing dataset...")
t = tsp.Tsp()
X, Y = t.next_batch(3)#10000
#X.shape
#(3, 10, 2)
#Y.shape
#(3, 10)
x_test, y_test = t.next_batch(2)
YY = []
for y in Y:
YY.append(to_categorical(y))
YY = np.asarray(YY)
hidden_size = 128
seq_len = 10
nb_epochs = 100
def runIt(hidden_size=128, learning_r=0.1, batch_s=30000, sameLen=True):
print("preparing dataset...")
t = tsp.Tsp()
if sameLen:
X, Y = t.next_batch(batch_s)
else:
X, Y = t.next_small_batch(batch_s)
x_test, y_test = t.next_batch(1)
learning_rate = learning_r
YY = []
for y in Y:
YY.append(to_categorical(y))
YY = np.asarray(YY)
X_val, Y_val = t.next_batch(1000)
YY_val = []
for y in Y_val:
YY_val.append(to_categorical(y))
YY_val = np.asarray(YY_val)
seq_len = 6
print("building model...")
main_input = Input(shape=(seq_len, 3), name='main_input')
encoder = LSTM(output_dim=hidden_size,
return_sequences=True,
name="encoder")(main_input)
decoder = PointerLSTM(hidden_size, output_dim=hidden_size,
name="decoder")(encoder)
model = Model(input=main_input, output=decoder)
model.compile(optimizer='adadelta',
loss='categorical_crossentropy',
metrics=['accuracy'])
history = LossHistory()
#X_val, Y_val = t.next_small_batch(1000)
model.fit(X,
YY,
nb_epoch=nb_epochs,
batch_size=64,
callbacks=[LearningRateScheduler(scheduler), history],
validation_data=(X_val, YY_val))
score_val = model.evaluate(X_val, YY_val, verbose=0)
print("Let's do one test!")
print("The Data I want to predict:")
print(x_test)
print("The result should be", y_test)
print(model.predict(x_test))
print("------")
print(to_categorical(y_test))
model.save_weights('model_weight_100.hdf5')
#score_train=model.evaluate(X, YY, verbose=0)
#print('Train score:', score_train[0])
#print('Train accuracy:', score_train[1])
print('Test score:', score_val[0])
print('Test accuracy:', score_val[1])
history.loss_plot('epoch')