Closed AntonioCarta closed 1 year ago
added deploy configuration to set the folders of each server in a single place.
It would also be nice to have a debug configuration that runs on a separate folder to avoid pulling the final experimental results with debug runs.
Yes, right now I override the results so that could be the risk, but I hate to fill my directories with useless results folders. I can add an option to append a "debug" folder inside the results maybe ? Like experiment.debug and if it's set to true I log into this folder.
Usually I have a debug
flag. If set to True, it switches to $RESULT_FOLDER/debug
. You don't even need to divide it by data/method, you can just clean it each time before starting.
Also, I often change other hyperparameters just to speed things up and help with debugging (num_epochs, subsampling data, ...).
Can you merge this change if it's ok for you?
Usually I have a
debug
flag. If set to True, it switches to$RESULT_FOLDER/debug
. You don't even need to divide it by data/method, you can just clean it each time before starting. Also, I often change other hyperparameters just to speed things up and help with debugging (num_epochs, subsampling data, ...).Can you merge this change if it's ok for you?
It's fine but I just don't understand the datadir changes you made. Previously I was giving only the root of the datasets but now you changed it I think it's not going to work.
added deploy configuration to set the folders of each server in a single place.
It would also be nice to have a debug configuration that runs on a separate folder to avoid pulling the final experimental results with debug runs.