Maxim Egorov Nova
Maxim Egorov Nova
I know that Reinforce.jl is not trying to emulate OpenAI gym exactly, but I'm curious behind the reasoning to a couple interface decisions that seem inconsistent with gym's. First, why...
Minor change to import path for `rnn_cell`. Not backwards compatible with versions below 0.9.
Currently, the solver only supports Gauss-Siedel value iteration. We should have an optional flag to do value iteration without the Gauss-Siedel method if the user wishes. There are a few...
After a large number of iterations HDF5 whiffs and throws the error: ``` python RuntimeError: Unable to register datatype (Can't insert duplicate key) ``` More details can be found here:...
We should allow the user to pick what datatype they want to use for training. We can add this to the settings dictionary. It could look like ``` python 'dtype'...
Does it make sense for the visualization to be on while experience tuples are being generated? We can just visualize during evaluation, since that will show how well the policy...
We should include tensorflow as a potential backend for the framework.
Should we automatically check if the user is requesting only the current observation as input to the DQN? We can then decide to use a more efficient memory sampling scheme...