Data isn't prepared correctly for invasive-species-monitoring

[rubzo]

The prepare function for the invasive-species-monitoring competition appears to incorrectly create .7z files using py7zr:

mle-bench/mlebench/competitions/invasive-species-monitoring/prepare.py

Line 88 in 1d391b0

with py7zr.SevenZipFile(public / "train.7z", "w") as z:

This was brought to my attention because an agent was unable to find any images in the train.7z or test.7z datasets.

You can repro the issue simply by creating a test case like:

import py7zr
import pathlib

def main():
    p = pathlib.Path("project")
    with py7zr.SevenZipFile(pathlib.Path("project.7z"), "w") as z:
        z.write(p)

And putting some data in a directory called project. The resulting project.7z file is invalid.

This is fixed by changing the use of 'write' to 'writeall' i.e:

z.writeall(p)

Then the .7z file is correctly created.

I also see that 'statoil-iceberg-classifier-challenge' is the only other competition that uses this package, but I haven't tested if it's also affected.

[joe-needham]

Thanks for reporting this. I've raised the PR #131 to add this to the known issue list in the README (presumably the agent will struggle with this task without the images, so to preserve the current leaderboard we want to record these issues, and then try and fix all such issues in batch at some point in future). Let me know if I misunderstood the issue.