/dev/null is seen as a file and thus fails. In the end I just have a clean up process at the end of my optimization run that deletes all ray_result subdirectories. I am a bit tight on space and with lots for walk forward optimization the disk fills up.
I am in a similar situation. There should be a way to disable the logging of the progress.csv file on ray_results? My understanding is that the progresss.csv file is not used by tensorboard or any other relevant process, so what I do is that I use the analysis object after tune.run() to save a custom .csv with fewer columns (the big issue is with the hist-stats column). Right now, I am erasing the progress.csv file manually, but it would be great to change the Logger object directly so the progress.csv file is not created from the get-go. I tried to parse the documentation on Custom Loggers but that seems to be tailored to adding information to loggers but not to take out the progress.csv file. Any help would be appreciated.