Strange behavior in the "tiger_antelope" example about ai-toolbox HOT 3 CLOSED

The warnings unfortunately do not depend from me: the Eigen library seems to be doing things with enums which do not play nicely with g++-7.0. They have an open bug report on this, and I hope they will fix it soon. If you actually check you'll see the warnings are always pretty much the same two things, inside the Eigen library.

It is definitely NOT ok that both the Python example nor the C++ version with a 5x5 grid are not working as expected. If the solving process has converged you should expect to see optimal behavior always (if the configuration of the problem was correct), so I'm going to check on this ASAP.

In general is not that Python is slower in solving the problem. It's mostly how the code creates the transition/reward functions ininitally from Python requires lots of loops, which Python dislikes, and so ends up taking a ton of time creating the model and importing it into C++. After that, it's instant.

I'll let you know when things are working properly again, thanks!

from ai-toolbox.

Hey, I think I found the problem.

I while ago I did some work transitioning from Boost data structures to Eigen, and forgot that Eigen does not automatically zero its matrices. This resulted in a policy with generally random values - it was luck that it worked at all in the normal case!

I've tested this now and it seems to work fine. Let me know if it works or if you have any more problems!

from ai-toolbox.

Yup, it seems to work fine now. Thanks!

from ai-toolbox.