def load_train_data(stock_id):
stock_df = stock_kline_day(stock_id, qfq)
values, labels = process_features(stock_df)
inputs_x, inputs_y = to_time_series(values, labels)
X_train, X_test, y_train, y_test = train_test_split(inputs_x,
inputs_y,
test_size=test_size,
random_state=seed)
return data_to_tensor(X_train, X_test, y_train, y_test)
def predict(model, stock_id):
stock_df = stock_kline_day(stock_id, qfq)
values, _ = process_features(stock_df)
index = len(values)
values = np.array([values[index - seq_length:index]])
values = torch.tensor(values).float().to(device=device)
with torch.no_grad():
outputs = model(values)
return outputs
def predict(model, stock_id):
stock_df = stock_kline_day(stock_id, qfq)
values, _ = process_features(stock_df)
index = len(values)
values = np.array([values[index - seq_length:index]])
values = torch.tensor(values).float().to(device=device)
with torch.no_grad():
outputs = model(values)
return pd.DataFrame(outputs.reshape(-1, 2).cpu().numpy(),
columns=['high', 'low'])
def generate_dataset(stock_ids):
x_list = []
y_list = []
for stock_id in stock_ids:
stock_df = stock_kline_day(stock_id, qfq)
values, labels, _ = process_features(stock_df)
inputs_x, inputs_y = to_time_series(values, labels)
x_list.append(inputs_x)
y_list.append(inputs_y)
c = list(zip(x_list, y_list))
random.shuffle(c)
x_list, y_list = zip(*c)
x = np.concatenate(x_list, axis=0)
y = np.concatenate(y_list, axis=0)
return x, y
def predict(model, stock_id):
model = load_pre_trained_model(model, stock_id)
data_context = load_data_context(stock_id)
model.eval()
stock_df = stock_kline_day(stock_id, qfq)
values, _ = process_features(stock_df, data_context)
index = len(values)
values = np.array([values[index - seq_length:index]])
values = torch.tensor(values).float().to(device=device)
with torch.no_grad():
outputs = model(values)
outputs = pd.DataFrame(outputs.reshape(-1, 2).cpu().numpy(),
columns=['high', 'low'])
for column in outputs.columns:
sc = data_context[column]
values = outputs.loc[:, [column]].values
values = sc.inverse_transform(values)
outputs[column] = pd.Series(values.reshape(-1))
return outputs
def load_train_data(stock_ids):
inputs_x_list = []
inputs_y_list = []
data_context = {}
for stock_id in stock_ids:
if not stock_id in data_context:
data_context[stock_id] = {}
stock_df = stock_kline_day(stock_id, qfq)
values, labels = process_features(stock_df, data_context[stock_id])
inputs_x, inputs_y = to_time_series(values, labels)
inputs_x_list.append(inputs_x)
inputs_y_list.append(inputs_y)
save_data_context(data_context)
input_x = np.concatenate(inputs_x_list, axis=0)
input_y = np.concatenate(inputs_y_list, axis=0)
X_train, X_test, y_train, y_test = train_test_split(input_x,
input_y,
test_size=test_size,
random_state=seed)
return data_to_tensor(X_train, X_test, y_train, y_test)
def persist_stock_info():
pbar_stock_ids = tqdm(default_stock_ids)
for stock_id in pbar_stock_ids:
stock_df = stock_kline_day(stock_id, qfq)
stock_df.to_csv(os.path.join(data_folder, "{0}.csv".format(stock_id)))