search

Ivsxk / RAT

61 stars 43 forks source link

How to reproduce the results on more recent data? #6

Open jnenx opened 3 years ago

jnenx commented 3 years ago

Hi,

I am trying to replicate the results of this algo but everytime I try I get very, very poor results. I am wondering how the data was preprocessed. Could you clarify that? Any help is very welcome.

Thanks!

Ivsxk commented 3 years ago

Hi,

I am trying to replicate the results of this algo but everytime I try I get very, very poor results. I am wondering how the data was preprocessed. Could you clarify that? Any help is very welcome.

Thanks!

To generate Data.db about the latest data, please refer to the code contributed by ZhengyaoJiang(https://github.com/zhengyaujiang/PGPortfolio). You can also write your own code to generate data files that meet the following rules： image

jnenx commented 3 years ago

Hi, thanks a lot for the reply. About this step described in the paper:

"we normalize the price series of each asset by element-wise dividing the prices regarding the last period in the price series."

Is that done by the code deposited here, or that is a previous step that I have to do before processing the dataset?

jnenx commented 3 years ago

You did mention to generate the .db file from the code contributed by ZhengyaoJiang. But is this .db file compatible with the code from RAT?

I managed to normalize the data with the pandas library however when I save it as .db file I am not able to preserve the information fully and that results in the error: ValueError: the length of selected coins 0 is not equal to expected 10. So the algorithm is not able to see the coin data as before which results in no coins to be found. I was not able to find a way to resolve this issue in particular with pandas.

jnenx commented 3 years ago

The link above is no longer working ZhengyaoJiang(https://github.com/zhengyaujiang/PGPortfolio)

Ivsxk commented 3 years ago

Hi, thanks a lot for the reply. About this step described in the paper:

"we normalize the price series of each asset by element-wise dividing the prices regarding the last period in the price series."

Is that done by the code deposited here, or that is a previous step that I have to do before processing the dataset?

The normalization is performed in the forward function of the EncoderDecoder: image The price series is a tensor with [4 kinds of prices, batch size, window size, number of coins]. price_series[0:1,:,-1:,:] is the close price of the last period in a window.

Ivsxk commented 3 years ago

You did mention to generate the .db file from the code contributed by ZhengyaoJiang. But is this .db file compatible with the code from RAT?

I managed to normalize the data with the pandas library however when I save it as .db file I am not able to preserve the information fully and that results in the error: ValueError: the length of selected coins 0 is not equal to expected 10. So the algorithm is not able to see the coin data as before which results in no coins to be found. I was not able to find a way to resolve this issue in particular with pandas.

It's compatible. You only need to use ZhengyaoJiang's code to generate Data.db file. The normalization is done in RAT.

jediyoda36 commented 3 years ago

@Ivsxk Same problem here... image

The link you mentioned above is no longer available

ZhengyaoJiang(https://github.com/zhengyaujiang/PGPortfolio).