Reassign pipe logger

[janosh]

When assigning custom loggers during instantiation of a MatPipe, i.e.

import automatminer as amm
import logging

my_logger = logging.getLogger("my_logger")

pipe = amm.MatPipe(logger=my_logger)

I somehow sometimes end up with pipe._logger == None. Haven't quite figured out when exactly that happens. And it only throws after training once I try to pipe.predict with

AttributeError: 'NoneType' object has no attribute 'info'

at self.logger.info. To proceed, I tried to reassign pipe._logger which didn't work at first until propagating that reassignment to all of pipe.(autofeaturizer|cleaner|reducer|learner).

Since you're already using @property for the logger, how about adding a setter to facilitate reassigning the logger?

[janosh]

I just isolated that

pipe.save("mat.pipe")

causes the logger to be removed. If I don't run that before

pipe.predict(df)

everything is fine. If I do, I get

AttributeError: 'NoneType' object has no attribute 'info'

at

self.logger.info("Beginning MatPipe prediction using fitted pipeline.")

[janosh]

Here's the problem. The loggers need to be reassigned after pickling.