Join GitHub today
GitHub is home to over 40 million developers working together to host and review code, manage projects, and build software together.Sign up
Precision Loss whe saving ACPolicy using RL4J #5543
I made an education of RL agent using the A3CDiscreteConv. Then I saved the policy in a file. I played the agent with the policy which I still hold in memory. After that I played the policy which I loaded from file. The results were very different and I had a message telling me I will loose precision:
So I didn't changed any of the default datatypes in global configuration. How can I make the model from the file to be equal to the one in memory.
Here is some code:
But I'm pretty sure
Anyway, the seed used for the network configuration is different from the one used by ACPolicy... Should we still have it use the seed from the network by default instead of not using randomness by default?