a['Close'].replace(0, np.nan, inplace=True)
a['Close'].fillna(method='ffill', inplace=True)
values = a['Close'].values.reshape(-1, 1)
values = values.astype('float32')
scaler = MinMaxScaler(feature_range=(0, 1))
scaled = scaler.fit_transform(values)
train_size = int(len(scaled) * 0.8)
valid_size = int(len(scaled) * 0.9)
test_size = len(scaled) - train_size
train, valid = scaled[0:train_size, :], scaled[train_size:valid_size, :]
test = scaled[valid_size:, :]
train_dict = create_data_dict(train, 4)
test_dict = create_data_dict(valid, 4)
model = LSTMRegressor(input_size=1,
hidden_size=32,
dropout_rate=0.5
)
predictor = Predictor(model=model,
train_data=train_dict,
test_data=test_dict,
batch_size=64,
use_cuda=True)
predictor.fit(300, 30)
results = predictor.predict(predictor.test_loader).reshape(1, -1)[0]
test_y = test_dict["y"].numpy().reshape(1, -1)[0]
pyplot.plot(results, label='predict')
pyplot.plot(test_y, label='true')
pyplot.legend()
pyplot.show()
train_size:train_size + valid_size, :]
test = scaled_features[train_size + valid_size:, :]
train_dict = create_data_dict(train, 4)
valid_dict = create_data_dict(valid, 4)
test_dict = create_data_dict(test, 2)
model = LSTMRegressor(input_size=len(columns),
hidden_size=32,
dropout_rate=0.5)
predictor = Predictor(model=model,
train_data=train_dict,
test_data=valid_dict,
batch_size=64,
use_cuda=True)
predictor.fit(400, 50)
results = predictor.predict(predictor.test_loader).reshape(1, -1)[0]
test_y = valid_dict["y"].numpy().reshape(1, -1)[0]
print(test_y)
axes = pyplot.gca()
new_results = results * data_range + data_min
new_test_y = test_y * data_range + data_min
initial_date = datetime(2018, 4, 15)
final_result = []
for quote, predict in zip(new_test_y, new_results):
initial_date = initial_date + timedelta(days=1)
date_str = initial_date.strftime("%Y-%m-%d")
final_result.append({
"date": date_str,
"quote": float(quote),
