def get_instrument_weights(self):
"""
Get the time series of instrument weights, accounting for potentially missing positions, and weights that don't add up.
:returns: TxK pd.DataFrame containing weights, columns are instrument names, T covers all subsystem positions
"""
self.log.terse("Calculating instrument weights")
if self.use_estimated_instrument_weights():
raw_instr_weights = self.get_raw_estimated_instrument_weights()
else:
raw_instr_weights = self.get_raw_fixed_instrument_weights()
instrument_list = list(raw_instr_weights.columns)
subsys_positions = [
self.get_subsystem_position(instrument_code)
for instrument_code in instrument_list
]
subsys_positions = pd.concat(subsys_positions, axis=1).ffill()
subsys_positions.columns = instrument_list
instrument_weights = fix_weights_vs_position_or_forecast(
raw_instr_weights, subsys_positions)
smooth_weighting = self.parent.config.instrument_weight_ewma_span
# smooth
instrument_weights = instrument_weights.ewm(smooth_weighting).mean()
return instrument_weights
def get_unsmoothed_instrument_weights_fitted_to_position_lengths(
self) -> pd.DataFrame:
raw_instrument_weights = self.get_unsmoothed_raw_instrument_weights()
instrument_list = list(raw_instrument_weights.columns)
subsystem_positions = [
self.get_subsystem_position(instrument_code)
for instrument_code in instrument_list
]
subsystem_positions = pd.concat(subsystem_positions, axis=1).ffill()
subsystem_positions.columns = instrument_list
## this should remove when have NAN's
## FIXME CHECK
instrument_weights = fix_weights_vs_position_or_forecast(
raw_instrument_weights, subsystem_positions)
# now on same frequency as positions
# Move to daily for space saving and so smoothing makes sense
daily_unsmoothed_instrument_weights = instrument_weights.resample(
"1B").mean()
return daily_unsmoothed_instrument_weights
def _fix_weights_to_forecasts(
self, instrument_code: str,
monthly_forecast_weights: pd.DataFrame) -> pd.DataFrame:
# we get the rule variations from forecast_weight columns, as if we've dropped
# expensive rules (when estimating) then get_trading_rules will give
# the wrong answer
rule_variation_list = list(monthly_forecast_weights.columns)
# time series may vary
forecasts = self.get_all_forecasts(instrument_code,
rule_variation_list)
# adjust weights for missing data
# also aligns them together with forecasts
forecast_weights_fixed_to_forecasts = fix_weights_vs_position_or_forecast(
monthly_forecast_weights, forecasts)
return forecast_weights_fixed_to_forecasts