search

rmattson1008 / dqn_asteroid_probe

Probing various layers of DQN to assess "difficulty" of decisions.
MIT License
0 stars 0 forks source link

Check Convs #1

Open rmattson1008 opened 1 year ago

rmattson1008 commented 1 year ago

I just set parameters to what works. Make sure the reasoning is solid, lines up with what other people do.

rmattson1008 commented 1 year ago

Then extend to dqn_h5 and dqn_h2

rmattson1008 commented 1 year ago

Am tempted to hook and flatten conv. as well. Don't. Keep conv as small as possible - only putting inputs into a state the dense layers can work on. Not a representation learning problem.