Danijar Hafner comments

Results 165 comments of


                                            Danijar Hafner

Failing to fully replicate Pong with A3C-LSTM

@miyosuda That video seems like the agent memorized the environment. I think the paper authors use random starts to create a fairer evaluation. They sample a number from 0 to...

Failing to fully replicate Pong with A3C-LSTM

I see, so the agent just learned a good behavior that results in very repetitive episodes.

Understanding normalization of advantage function

Doesn't subtracting the mean from the advantages have the effect of an entropy regularizer? Ignoring the clipping, the objective is `logp * (adv - mean) / std = logp *...

tmux-copycat slow on my machine

What's the reason for dumping the buffer to a text file before searching it? I'd be happy to help make this faster. The plugin is a great idea but right...

tmux-copycat slow on my machine

Would it be possible to bind the keys (maybe different ones to avoid conflicts) once during startup rather than every time copy mode is entered?

Shouldn't Ctrl+Return switch from command mode to Jupyter mode?

I see how it makes sense for code cells that Ctrl+Return runs the cell and stays in Vim command mode. But for markdown cells, the command doesn't do anything right...

Make Metrics thread-safe for use in A3C

Still occurs sometimes: `(sqlite3.OperationalError) database is locked`

Flights delay data set

Please email me with requests for the flights dataset. I'm not sure if we're allowed to publish the pre-processed dataset.

Do the V5 environments perform frame pooling?

Thanks for the detailed comment! I agree that averaging is nicer than taking the maximum, but at this point it's more important to be compatible with the vast existing literature...

Do the V5 environments perform frame pooling?

Hi @JesseFarebro, do you have an idea for a workaround here? It's the only issue holding me back from switching to `ale-py` and the new V5 envs. To implement max...