batch_size=batchsize, epochs=epochs, validation_data=(x_val, y_val))
# Output training history to csv
history_df = pd.DataFrame(history.history)
history_df.to_csv("training_history.csv")
print("Predicting...")
result = model.predict(x_test)
predicted = pd.DataFrame(result)
predicted.columns = ['predicted_nikkei']
predicted['actual_nikkei'] = y_test
print("Completed Prediction.")
print(predicted.shape)
print(predicted[:10])
# Output to csv
predicted.to_csv("predicted.csv")
# Evaluate
evaluation_score = model.evaluate(x_test, y_test, batch_size=batchsize, verbose=1)
print("Evaluation score is {}".format(evaluation_score))
# Evaluate benchmark
dummy_predicted = DummyPredictor().get_evaluation_score(y_train)
dummy_evaluation_score = np.mean((y_test - dummy_predicted) ** 2)
print("Dummy evaluation score is {}".format(dummy_evaluation_score))
# Compare
comparison = (evaluation_score / dummy_evaluation_score) * 100
print("This prediction model's MSE is {} percent compared to benchmark. (smaller is better)".format(comparison))
sys.dont_write_bytecode = True
from data import DataLoader
from lstm_model import LSTMModel
from feed_forward_model import FeedForwardModel
k = 5
results = []
total = 0
for i in range(k):
dataLoader = DataLoader(src="MNIST_data/dataset")
[
noisyTrain, noisySymbolsTrain, targetsTrain, noisyValidation,
noisySymbolsValidation, targetsValidation, noisyTest, noisySymbolsTest,
targetsTest
] = dataLoader.getSequencialData(i, 2, 2)
model = LSTMModel(2, [512, 512], 1568, 20)
model.train(noisySymbolsTrain, targetsTrain, noisySymbolsValidation,
targetsValidation, 100, 200)
result = model.evaluate(noisySymbolsTest, targetsTest)
results.append(result[1])
total += result[1]
print(results)
print("SCORE: " + str(total / k))
[
noisyTrain,
noisySymbolsTrain,
targetsTrain,
noisyValidation,
noisySymbolsValidation,
targetsValidation,
noisyTest,
noisySymbolsTest,
targetsTest
] = dataLoader.getSequencialDataWithDontCare(0, 2, 2)
model = LSTMModel(2, [512, 512, 512], 1568, 20)
model.train(noisyTrain, targetsTrain, noisyValidation, targetsValidation, 100, 100)
result = model.evaluate(noisyTest, targetsTest)
predictions = model.predict(noisyTest)
count = 0
for index in range(len(predictions)):
a = utils.argmax(predictions[index][:10])
b = utils.argmax(predictions[index][10:])
predicted = a * 10 + b
a = utils.argmax(targetsTest[index][:10])
b = utils.argmax(targetsTest[index][10:])
actual = a * 10 + b
if actual == predicted:
count += 1