Deep reinforcement learning running locally - different results

Hi I'm finding the RL model not producing the same returns when running example solutions on my local jupyter notebook, maybe somebody can tell me if I'm missing something…



I have run the backtests on 'Apply RL on Mixed Pattern Wave' and 'RL_Model_on_Price_Data' and both of these have given worse returns on the default hyperparameters. I've adjusted test_mode = false, have I missed something?



Also in terms of increasing the backtesting speed of RL does anyone have experience running models with Azure, were you able to achieve a significant increase in training speed at a resonable cost for this size of problem?



Thank you

Hi Joseph,



The difference in returns is due to the presence of a random action generator in the code. So basically, there is a parameter EPSILON in the rl_config dictionary, which is the exploration/exploitation threshold. When the model is run, a random number is generated between 0 and 1. If the generated number is less than or equal to EPSILON, then the course of action would be exploration and hence a random action (buy, sell, hold) is selected randomly. As this randomness is involved in every run of the code, the strategy returns might vary.



Hope this helps!



Thanks,

Akshay