Reproduce AAAI22-FactorVAE

Known Issues

1. In factor_encoder.py, the PortfolioLayer misrepresent the paper's idea of "a portfolio set". The current version only use "one portfolio" instead of "a set of".
2. Dataset now is not consistent with the paper due to some personal reason.
3. In feature_extractor.py, the latent feature, which is the GRU's output, seems quirky.