How to adapt LSTM to DDG-DA

[Xxiaoting]

I want to adapt LSTM to DDG-DA, how can I do that?

What I have tried:

1. modify rolling_benchmark.py to fit with LSTM parameters
2. modify the bug caused by changing the dataset object to TSDatasetH. change the file in qlib > contrib > meta > data_selection > model.py:

    def reweight(self, data: Union[pd.DataFrame, pd.Series]):
        # TODO: handling TSDataSampler
        if isinstance(data, pd.DataFrame):
            idx = data.index
        else:
            idx = data.get_index()

        w_s = pd.Series(1.0, index=idx)
        for k, w in self.time_weight.items():
            w_s.loc[slice(*k)] = w
        logger.info(f"Reweighting result: {w_s}")
        return w_s

However, the valid loss remains the same in different epoch and I don't know why.

[you-n-g]

@Xxiaoting Do the data weights change here?
You can set a breakpoint here to debug.

[github-actions[bot]]

This issue is stale because it has been open for three months with no activity. Remove the stale label or comment on the issue otherwise this will be closed in 5 days