def predict_by_model(ref_date: str,
alpha_model: ModelBase,
data_meta: DataMeta = None,
x_values: pd.DataFrame = None,
codes: Iterable[int] = None):
if x_values is None:
predict_data = data_meta.fetch_predict_data(ref_date, alpha_model)
codes, x_values = predict_data['predict']['code'], predict_data['predict']['x']
return pd.DataFrame(alpha_model.predict(x_values).flatten(), index=codes), x_values
def train_model(ref_date: str,
alpha_model: ModelBase,
data_meta: DataMeta = None,
x_values: pd.DataFrame = None,
y_values: pd.DataFrame = None):
base_model = copy.deepcopy(alpha_model)
if not isinstance(alpha_model, ConstLinearModel):
if x_values is None:
train_data = data_meta.fetch_train_data(ref_date, alpha_model)
x_values = train_data['train']['x']
y_values = train_data['train']['y']
base_model.fit(x_values, y_values)
return base_model, x_values, y_values
fit_intercept = True
kernal_feature = 'roe_q'
regress_features = {kernal_feature: LAST(kernal_feature),
kernal_feature + '_l1': SHIFT(kernal_feature, 1),
kernal_feature + '_l2': SHIFT(kernal_feature, 2),
kernal_feature + '_l3': SHIFT(kernal_feature, 3)
}
const_features = {kernal_feature: LAST(kernal_feature)}
fit_target = [kernal_feature]
data_meta = DataMeta(freq=freq,
universe=universe,
batch=batch,
neutralized_risk=neutralized_risk,
risk_model=risk_model,
pre_process=pre_process,
post_process=post_process,
warm_start=warm_start,
data_source=data_source)
alpha_model = LinearRegression(features=regress_features, fit_intercept=True, fit_target=fit_target)
composer = Composer(alpha_model=alpha_model, data_meta=data_meta)
start_date = '2014-01-01'
end_date = '2016-01-01'
regression_model = LinearRegression(features=regress_features, fit_intercept=fit_intercept, fit_target=fit_target)
regression_composer = Composer(alpha_model=regression_model, data_meta=data_meta)
data_package1 = fetch_data_package(engine,
}
weights = {'f1': 1., 'f2': 0.}
# alpha_model = XGBTrainer(objective='reg:linear',
# booster='gbtree',
# n_estimators=300,
# eval_sample=0.25,
# features=alpha_factors)
alpha_model = ConstLinearModel(features=alpha_factors, weights=weights)
data_meta = DataMeta(freq=freq,
universe=universe,
batch=1,
neutralized_risk=neutralized_risk,
pre_process=None,
post_process=None,
warm_start=1)
industries = industry_list('sw_adj', 1)
total_risk_names = ['total', 'benchmark'] + industries
b_type = []
l_val = []
u_val = []
for name in total_risk_names:
if name == 'total':
b_type.append(BoundaryType.ABSOLUTE)
}
weights = {str(factor): 1.}
# alpha_model = XGBTrainer(objective='reg:linear',
# booster='gbtree',
# n_estimators=300,
# eval_sample=0.25,
# features=alpha_factors)
alpha_model = ConstLinearModel(features=alpha_factors, weights=weights)
data_meta = DataMeta(
freq=freq,
universe=universe,
batch=32,
neutralized_risk=None, # industry_styles,
pre_process=None, # [winsorize_normal, standardize],
post_process=None,
warm_start=12) # [standardize])
industries = industry_list('sw_adj', 1)
total_risk_names = ['total']
b_type = []
l_val = []
u_val = []
for name in total_risk_names:
if name == 'total':
b_type.append(BoundaryType.ABSOLUTE)
start_date = '2011-01-01'
end_date = '2011-05-04'
freq = '10b'
neutralized_risk = None
alpha_factors = {'ep_q_cs': CSQuantiles(LAST('ep_q'), groups='sw1_adj')}
weights = dict(ep_q_cs=1.)
alpha_model = ConstLinearModel(features=alpha_factors, weights=weights)
data_meta = DataMeta(freq=freq,
universe=universe,
batch=1,
neutralized_risk=None,
pre_process=None,
post_process=None,
data_source=os.environ['DB_URI'])
strategy = Strategy(alpha_model,
data_meta,
universe=universe,
start_date=start_date,
end_date=end_date,
freq=freq,
benchmark=benchmark_code)
strategy.prepare_backtest_data()
def create_scenario(weights_bandwidth=0.02,