Write out cross-validation results immediately

[LukasBeiske]

Fixes #2480

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Comparison is base (75a38ee) 92.44% compared to head (19791c9) 92.47%. Report is 21 commits behind head on main.

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #2483 +/- ## ========================================== + Coverage 92.44% 92.47% +0.02% ========================================== Files 234 234 Lines 19917 19965 +48 ========================================== + Hits 18413 18462 +49 + Misses 1504 1503 -1 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[LukasBeiske]

Good start, but you should

a) Check overwrite conditions in __init__: if output file exists and not overwrite, raise an error, if output file exists and overwrite=true truncate it (tables.open_file(path, mode='w').

b) Move the writing into the cv loop using write_table(..., append=True) to further reduce memory usage

c) Add writing of metadata to the output file

a) The error if file exists and not overwrite is already raised by the train tools via Tool.check_output.

a)/b) Using write_table(..., append=True, mode='w') does not work, does it? Isn't this basically the same as write_table(..., append=True, overwrite=True)?

c) Is there any other metadata requiered besides some column descriptions?

[maxnoe]

a) The error if file exists and not overwrite is already raised by the train tools via Tool.check_output.

Also for the cv output file and not just the model?

[maxnoe]

a)/b) Using write_table(..., append=True, mode='w') does not work, does it? Isn't this basically the same as write_table(..., append=True, overwrite=True)?

You need to truncate the file once in __init__ and then only use append=True

[LukasBeiske]

a) The error if file exists and not overwrite is already raised by the train tools via Tool.check_output.

Also for the cv output file and not just the model?

Yes, both output paths are passed and CrossValidator.ouput_path gets ignored if it is None