Why don't Notebook results match paper results - NeurIPS 2018

[benstaf]

The NeurIPS 2018 paper Practical Deep Reinforcement Learning Approach for Stock Trading announces that RL is better that DJIA index:

However, the notebook doesn't reproduce the result: https://github.com/AI4Finance-Foundation/FinRL/blob/master/examples/Stock_NeurIPS2018_3_Backtest.ipynb

Why?

[krishdotn1]

Hello @benstaf I think research paper result is trained on different time period for dji and notebook mentioned in the GitHub is trained for different period, can you match the period and hyperparameter as discussed in paper. You might see similar results.

Let me know if I did wrong interpretation of your issue or any thing else.

[benstaf]

Thanks, but in the paper I didn't find the hyperparameters used

[aymeric75]

the question is, why trying to make finrl works when FinRL_Podracer is in da place !