openbrain
openbrain copied to clipboard
mlberkeley
There are no commits, but it would be nice if we were able to read the white paper. Maybe convert it to a pdf and put it in the repo?
The ShareLaTeX Cloud Compiler was shut down in July, and as a result the docs (and whitepaper) are essentially unreadable. I do not know of another online LaTeX compiler, but...
Quick link to files in PR https://github.com/mlberkeley/openbrain/pull/14/files
Need to be able to do batched conditionals in tensorflow. At the current moment we aren't calculating gamma loss with the reward function. Add a replay_memory to the subcritic network...
Compse into an array. Do the same with all of the sub-critics, then compose into array. Store the data.
- Confirm that Q values do not diverge - Make sure environments are non-linear. (Linear updates still work)
Calculate the L_{1 or 2} norm between the two values.
- Base it off of the DDPG robustness test - Make sure that our network parameters are appropriate (20 neurons is probably only applicable to the mountain car enviro)
