quantlab.model.predefined.membership_mask.MembershipMaskedPredictor#

class quantlab.model.predefined.membership_mask.MembershipMaskedPredictor(predictor, membership)[source]#

Bases: PredictorView

A Predictor whose predictions are NaN where the symbol is not an index member.

predict_window asks the wrapped predictor for its predictions and keeps a cell only when the membership panel’s is_member is true on that bar’s date for that symbol; every other protocol member forwards to the wrapped predictor unchanged (PredictorView). The mask at bar t reads the membership of t’s date and nothing later, so it is information at t. Prices, tradability and delistings are untouched: a security that leaves the index keeps its prices, its prediction turns NaN, and the portfolio rule decides what happens to a holding (TopN sells it at the next rebalance, mean-variance holds it at mu=0).

A symbol missing from the membership panel is not a member. A bar whose date the membership panel does not cover (before its first or after its last day) is refused, because unknown membership is not “not a member”.

Parameters:

predictorPredictor: The model or ensemble whose predictions are masked.
membershipIndexConstituentDataset: The point-in-time membership, with a stored is_member panel on the symbol axis of the predictions (PERMNOs for CRSP stores).

Attributes:

predictorPredictor: The wrapped predictor.
membershipIndexConstituentDataset: The membership dataset.

Examples

With model a model over CRSP PERMNOs and membership the CrspSP500ConstituentDataset over the same roster:

>>> masked = MembershipMaskedPredictor(model, membership)
>>> backtester = USEquityCrossectionSelectStockVectorBt(
...     CrossSectionBacktestConfig(
...         price_dataset=crsp_index_dataset,  # unmasked prices
...         model=masked, model_mode="load", checkpoint=checkpoint,
...         start_date="2020-01-01", end_date="2024-12-31",
...         output_dir="runs", rebalance_periods=5,
...         constructor=TopNConstructor(TopNConfig(top_n=50)),
...     )
... )
>>> result = backtester.run()  # predictions.zarr is masked too

config_cls#

The config dataclass the wrapper is serialised and rebuilt with.

alias of MembershipMaskConfig