- 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". - Dataset now is not consistent with the paper due to some personal reason.
- In
feature_extractor.py
, the latent feature, which is the GRU's output, seems quirky.
ytliu74 / factorvae Goto Github PK
View Code? Open in Web Editor NEWReproduce AAAI22-FactorVAE