Fix736

[PGijsbers]

Fixes #736.

• All openml package logging is logged through the 'openml' logger (this fixes side-effects of working with the root logger that are demonstrated in the issue and code below).
• All messages at logging.WARNING and up are printed to console (by default). This is in line with old behavior.
• A file log of messages at logging.DEBUG and up (i.e. all of them) are logged to a file in the cache directory (by default). A new log file is created if the old log file reaches a size of 1 MB. At most one current log file and and old log file will be kept on disk at a time (i.e. at most 2 MB of openml-python logs are kept).
• The log level defaults can be changed through the configuration file, for both channels individually. A new file_verbosity is introduced alongside the regular verbosity.

The difference can be seen when running the following lines of code:

import logging

if __name__ == '__main__':
    log = logging.getLogger('mypackage')
    log.setLevel(logging.DEBUG)

    log.info("Message one")
    import openml
    log.info("Message two") 

    from openml.flows import get_flow
    f = get_flow(123)
    f.description = ''
    f._to_dict()  # logging.warning will be called due to empty description.

    from openml.datasets import get_dataset
    iris = get_dataset(61)  # since I have it in cache, a logging.debug statement is called.

in current develop we will see:

[INFO] [14:11:03:mypackage] Message two
[DEBUG] [14:11:03:mypackage] Message three
[WARNING] [14:11:03:openml.flows.flow] Flow 'weka.LinearRegression's empty description

printed in console, with no (semi-)permanent of the log is kept automatically.

With this PR, the console output becomes:

[WARNING] [14:12:42:openml.flows.flow] Flow 'weka.LinearRegression's empty description

and in my cache directory I find a file 'openml_python.log' with the following content:

[WARNING] [14:12:59:openml.flows.flow] Flow 'weka.LinearRegression's empty description
[DEBUG] [14:12:59:openml.datasets.dataset] Data pickle file already exists and is up to date.

I don't know how to unit test this, if at all.

---

A note on examples. I noticed some examples use logging (run setup and the fanova example). The generated pages do not show any logging output, so I was wondering what the purpose for the logging is. Perhaps @janvanrijn can help me here, as he seems to be the author.

[mfeurer]

The generated pages do not show any logging output, so I was wondering what the purpose for the logging is. Perhaps @janvanrijn can help me here, as he seems to be the author.

I guess we should move to print then. Could you please change that?

[mfeurer]

This looks good besides my (small comments). Could you please also add this to the changelog?

[codecov-io]

Codecov Report

Merging #861 into develop will decrease coverage by 0.04%.
The diff coverage is 82.92%.

@@             Coverage Diff             @@
##           develop     #861      +/-   ##
===========================================
- Coverage    88.43%   88.38%   -0.05%     
===========================================
  Files           37       37              
  Lines         4272     4298      +26     
===========================================
+ Hits          3778     3799      +21     
- Misses         494      499       +5

Impacted Files	Coverage Δ
openml/datasets/dataset.py	87.03% <100%> (ø)	⬆️
openml/extensions/sklearn/extension.py	91.32% <70%> (+0.01%)	⬆️
openml/config.py	90.47% <86.2%> (+0.47%)	⬆️
openml/_api_calls.py	82.14% <0%> (-3.58%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4a13100...f22f951. Read the comment docs.

[mfeurer]

I know it wasn't part of my original requests, but what do you think about having an example in which we show how one can easily change the log level? That might be good for folks who do ML, but do not care too much about how logging works in python.