def run_prediction(params: PredictParams):
try:
date_range = pd.date_range(parser.parse(params.end_date) +
datetime.timedelta(days=1),
params.forecast_date,
freq=params.offset.value)
model = getattr(source.back.models,
params.model.value).Model(params.params)
model.load(params)
res_y = model.train_and_predict(len(date_range))
res_index = list(date_range)
if params.upload:
real_df = make_df(params.uploaded_data, params.start_date,
params.forecast_date)
else:
real_df = DataProcess.load_data_from_moex(params.ticker,
params.start_date,
params.forecast_date,
params.offset.value)
except:
print(traceback.format_exc())
return PredictionData(status=cfg.Status.fail,
data=cfg.PREDICTION_FAILED)
return PredictionData(data={
"X": list(dates_from_array(real_df.index)),
"Y": list(real_df[params.ticker]),
"PredictedX": list(dates_from_array(res_index)),
"PredictedY": list(res_y)
},
status=cfg.Status.ready)
def load(self, params: PredictParams):
if params.upload:
loaded_df = make_df(params.uploaded_data, params.start_date, params.end_date)
else:
loaded_df = DataProcess.load_data_from_moex(params.ticker, params.start_date, params.end_date,
params.offset.value, self.exogenous_variables)
self.df = DataProcess.get_prepared_data_frame(loaded_df, predict_day=0)
def load(self, params: PredictParams):
if params.upload:
loaded_df = make_df(params.uploaded_data, params.start_date,
params.end_date)
else:
loaded_df = DataProcess.load_data_from_moex(
params.ticker, params.start_date, params.end_date,
params.offset.value)
self.df = loaded_df[loaded_df.columns[0]].reset_index(drop=True)
def run_cross_validation(params: PredictParams):
avg_mse = 0
avg_mape = 0
try:
if params.upload:
loaded_df = make_df(params.uploaded_data, params.start_date,
params.end_date)
else:
loaded_df = DataProcess.load_data_from_moex(
params.ticker, params.start_date, params.end_date,
params.offset.value,
params.params.get('exogenous_variables', []))
mse = []
mape = []
for i in tqdm(
range(
0, loaded_df.shape[0] - params.cv_period -
params.cv_predict_days, params.cv_shift)):
model = getattr(source.back.models,
params.model.value).Model(params.params)
local_params = copy.deepcopy(params)
local_params.start_date, local_params.end_date = loaded_df.index[
i], loaded_df.index[i + params.cv_period - 1]
local_params.upload, local_params.uploaded_data = params.upload, params.uploaded_data
model.load(local_params)
res = [[], []]
res[0] = model.train_and_predict(params.cv_predict_days)
res[1] = list(
loaded_df.iloc[i + params.cv_period:i + params.cv_period +
params.cv_predict_days, 0])
mse.append(metrics.mean_squared_error(res[1], res[0]))
mape.append(metrics.mean_absolute_percentage_error(res[1], res[0]))
if mse[-1] > 18000:
print(
mse[-1], loaded_df.index[i],
loaded_df.index[i + params.cv_period - 1],
loaded_df.index[i + params.cv_period +
params.cv_predict_days - 1])
avg_mse = sum(mse) / len(mse)
avg_mape = sum(mape) / len(mape)
except:
print(traceback.format_exc())
return PredictionData(status=cfg.Status.fail,
data=cfg.CROSS_VALIDATION_FAILED)
return PredictionData(data={
"mse": avg_mse,
"mape": avg_mape
},
status=cfg.Status.ready)