def _concat_factor_data(self, factor):
py_assert(self.freq == factor.freq,
ValueError,
'Failed to concatenate: factor {0} has different freq from containter'.format(factor.name))
self._check_tiaocang_date(factor)
self.data = pd.concat([self.data, factor.data], axis=1)
def remove_factor(self, factor):
factor_name = factor if isinstance(factor, str) else factor.name
py_assert(factor_name in self.names, ValueError,
'unable to remove factor_name {0}, which does not exist in current container'.format(factor_name))
self.names.remove(factor_name)
self.data.drop(factor_name, axis=1, inplace=True)
self.property = dissoc(self.property, factor_name)
return
def _check_tiaocang_date(self, factor):
"""
如果该因子的freq和container的freq一致，那么该因子的日期列应该包含所有的调仓日期，否则报错
"""
py_assert(set(self._tiaocang_date).issubset(set(factor.trade_date_list)),
ValueError,
'factor {0} does not contain all tiaocang date in its trade date list'.format(factor.name))
return
def _concat_factor_data(self, factor):
py_assert(
self.freq == factor.freq, ValueError,
'Failed to concatenate: factor {0} has different freq from containter'
.format(factor.name))
self._check_tiaocang_date(factor)
self.data = pd.concat([self.data, factor.data], axis=1)
def fit(self, factor_container, **kwargs):
self.factor_name = factor_container.alpha_factor_col if self.factor_name is None else self.factor_name
py_assert(set(self.factor_name).issubset(set(factor_container.alpha_factor_col)), ValueError,
'factor_name must be one of alpha factors in factor container')
nb_factor = len(self.factor_name)
self.weight = [1.0 / nb_factor] * nb_factor if self.weight is None else self.weight
self.ascend_order = [1] * nb_factor if self.ascend_order is None else self.ascend_order
return self
def get_industry_weight_on_date(self, date):
py_assert(self.industry_weight is not None, ValueError, 'industry weight data is empty')
date = ensure_datetime(date)
data = self.industry_weight.loc[date]
data = data.reset_index().set_index(INDEX_INDUSTRY_WEIGHT.industry_index)
data = data[data.columns[0]]
resid_weight = 100 - np.sum(data.values)
data['other'] = resid_weight if resid_weight > 0 else 0.0
return data
def top(df, column=None, n=5):
if isinstance(df, pd.Series):
ret = df.sort_values(ascending=False)[:n]
else:
py_assert(column is not None,
"Specify the col name or use pandas Series type of data")
ret = df.sort_values(by=column, ascending=False)[:n]
return ret
def remove_factor(self, factor):
factor_name = factor if isinstance(factor, str) else factor.name
py_assert(
factor_name in self.names, ValueError,
'unable to remove factor_name {0}, which does not exist in current container'
.format(factor_name))
self.names.remove(factor_name)
self.data.drop(factor_name, axis=1, inplace=True)
self.property = dissoc(self.property, factor_name)
return
def _check_tiaocang_date(self, factor):
"""
如果该因子的freq和container的freq一致，那么该因子的日期列应该包含所有的调仓日期，否则报错
"""
py_assert(
set(self._tiaocang_date).issubset(set(factor.trade_date_list)),
ValueError,
'factor {0} does not contain all tiaocang date in its trade date list'
.format(factor.name))
return
def get_industry_weight_on_date(self, date):
py_assert(self.industry_weight is not None, ValueError,
'industry weight data is empty')
date = ensure_datetime(date)
data = self.industry_weight.loc[date]
data = data.reset_index().set_index(
INDEX_INDUSTRY_WEIGHT.industry_index)
data = data[data.columns[0]]
resid_weight = 100 - np.sum(data.values)
data['other'] = resid_weight if resid_weight > 0 else 0.0
return data
def _merge_factors(self, factors):
if factors is None:
return
factors_ = factors if isinstance(factors, list) else list(factors)
for factor in factors_:
py_assert(factor.production_format == OutputDataFormat.MULTI_INDEX_DF,
ValueError,
'factor {0} does not in multi-index dataframe format therefore can not be merged into container'
.format(factor.name))
self._merge_factor(factor)
self.data = self.data.loc[self._tiaocang_date]
return
def fit(self, factor_container, **kwargs):
self.factor_name = factor_container.alpha_factor_col if self.factor_name is None else self.factor_name
py_assert(
set(self.factor_name).issubset(
set(factor_container.alpha_factor_col)), ValueError,
'factor_name must be one of alpha factors in factor container')
nb_factor = len(self.factor_name)
self.weight = [1.0 / nb_factor
] * nb_factor if self.weight is None else self.weight
self.ascend_order = [
1
] * nb_factor if self.ascend_order is None else self.ascend_order
return self
def _merge_factors(self, factors):
if factors is None:
return
factors_ = factors if isinstance(factors, list) else list(factors)
for factor in factors_:
py_assert(
factor.production_format == OutputDataFormat.MULTI_INDEX_DF,
ValueError,
'factor {0} does not in multi-index dataframe format therefore can not be merged into container'
.format(factor.name))
self._merge_factor(factor)
self.data = self.data.loc[self._tiaocang_date]
return
def _build_mapper(self, factor_container):
data_mapper_by_date = pd.Series()
industry_code = factor_container.industry_code
score = factor_container.score
for date in factor_container.tiaocang_date:
if self.method == SelectionMethod.INDUSTRY_NEUTRAL:
py_assert(self.industry_weight is not None, ValueError, 'industry weight has not been given')
industry_weight = self.industry_weight.loc[date]
data_mapper = [([score.name, industry_code.name],
IndustryNeutralSelector(industry_weight=industry_weight,
prop_select=self.prop_select,
min_select_per_industry=self.min_select_per_industry,
reset_index=True))]
else:
data_mapper = [(score.name, BrutalSelector(nb_select=self.nb_select,
prop_select=self.prop_select,
reset_index=True))]
data_mapper_by_date[date] = DataFrameMapper(data_mapper, input_df=True, df_out=True)
return data_mapper_by_date
def test_py_assert(self):
with self.assertRaises(ValueError):
py_assert(1 == 2, ValueError)