[FEAT]How to feed TimeGAN with different input/output array sizes

[amberyic]

I have many groups of sequence data, such as the stock data, I want to train GOOGL,AAPL,MSFT... in one model，but GOOGL/AAPL/MSFT have different sequence length. I hope the output can generate a similar variable length sequence. I want to know how to use this package to solve this type of problem.

[Salbinus]

Hi, the sequence length modeled by TimeGAN is not the complete history of the stock. It's rather a window of e.g. 24 time steps. So, you get around different length history by using the earliest date all stocks have in common.

[soya-miruku]

Hey, great work, how would you be able to say to look further than just 1 a time step, i.e. a horizon of say 2

[pbezz1]

@amberyic how did you solve the problem you reported on Github related to TimeGAN with different input/output array sizes. I'm interested to know because I'm facing the same problem. Would really appreciate your help. Thanks

[amberyic]

I tried to expand the variable-length sequence into a fixed-length sequence, but the effect was not good. Finally I used another model: https://github.com/fjxmlzn/DoppelGANger

@amberyic how did you solve the problem you reported on Github related to TimeGAN with different input/output array sizes. I'm interested to know because I'm facing the same problem. Would really appreciate your help. Thanks

[pbezz1]

@amberyic thanks for your reply. Can you give me some tips how you generated the data features for the DoppelGANger? Did you use stock prices or returns?