random

TFT predict() with output_dir() running out of memory on large validation sets

2

It appears there is some sort of memory leak. `new_raw_predictions = model.predict(validation, mode="prediction", return_index = True, batch_size=128, output_dir = PATH)`

Quote tick subscriptions are not always necessary (or available) yet features such as net exposures and unrealized_pnl depend on either quote or trade tick subscriptions to function. This should be...

MultiPeriodOptimization Behavior Documentation

15

With user provided returns forecast, with a single forecasted return per timestamp- how does the model use the forecast? Is it looking at the next `planning_horizon` bars of forecast on...

Feature request: handle user-defined time-varying universes (and better error checks with temporary `nan`s in user-provided returns)

13

I get the error: `ValueError: Parameter value must be real.` with callback to: ``` in FactorModelCovariance.values_in_time(self, **kwargs) sigma_sqrt = self.Sigma.current_value \ if self._alreadyfactorized \ else project_on_psd_cone_and_factorize( self.Sigma.current_value) # numpy eigendecomposition...

Questions on accounting model (e.g., for short positions)

3

Why is there the check: ``` v = np.sum(h) if v < 0.: raise DataError( f"Holdings provided to {self.__class__.__name__}.execute " + " have negative sum.") ``` In the case where...

Support for custom objective in lightgbm and xgboost

1

This is already supported in FLAML. See here: https://microsoft.github.io/FLAML/docs/Examples/AutoML-for-LightGBM