Hello,
I am configuring a tuner with a custom logger class that I derived from tune.logger.LoggerCallback.
I set it up that way:
myLogger=MyLoggerClass(someData)
tuner = tune.Tuner(
trainable=myTrainable,
run_config=train.RunConfig(callbacks=[myLogger]),
)
results=tuner.fit()
myLogger.DoSomeMoreStuff()
After I’m done with the tuning, I want to do some other stuff with myLogger class instance as shown above. All this works fine. But, then I want to introduce capability to restore a tune run. In the restore case, then I execute the following code:
tuner = tune.Tuner.restore(
exp_dir, trainable=myTrainable, resume_errored=True
)
tuner.fit()
When restoring, myLogger class will get also restored to a consistent state by rayTune. My problem is how can I get back a handle on myLogger class in the case of restoring so I can do some stuff with it after I’m done with the tuning, in the same way as what I was doing when not restoring ?
Thanks,
Edouard