·
10 commits
to main
since this release
Includes a major fix to random split initialization.
Now, CryptoEnvironment.evaluate() and CryptoEnvironment.reset() methods correctly implement random position splitting at reset and at the beginning of an evaluation period.
The purpose of random splitting is to ensure the agent can't avoid market movements by refusing to take a position. This is a reward-gaming behavior common to agents operating mis-specified trading environments.