search

ZhengyaoJiang / PGPortfolio

PGPortfolio: Policy Gradient Portfolio, the source code of "A Deep Reinforcement Learning Framework for the Financial Portfolio Management Problem"(https://arxiv.org/pdf/1706.10059.pdf).
GNU General Public License v3.0
1.74k stars 750 forks source link

Experience related to selected coins. #31

Closed AhmMontasser closed 6 years ago

AhmMontasser commented 6 years ago

Hello,

I noticed something while backtesting, the model new experience index is not following the previous backtesting if the selected coins is different.

does that mean that model experience gained is only related to the coins it's trained on ?

ZhengyaoJiang commented 6 years ago

I didn't clearly understand your question. What's the difference between “backtesting” and "previous backtesting"?

AhmMontasser commented 6 years ago

sorry for confusion, i meant that doing backtesting on different periods doesn't lead to the same sequence of experience index printed on this line DEBUG:root:a new experience, indexed by 5283, was appended

for example, backtesting from 1/ 4 /2017 to 1/7/2017 has an experience index sequence different from backtesting from 1/8/2017 to 1/12/2017.

my question: do the rolling training experience depends on the specific coins it was trained on ?

ZhengyaoJiang commented 6 years ago

Thanks for the clarification. The rolling train index could depend on many things, say, the batch size, window size and the number of training samples. One of the possible reasons here might be the dates represented by the string. I'm not sure if the types of coins would affect.

AhmMontasser commented 6 years ago

Thank you for explaining