ML rules fixes, new rule for msgpack-numpy

[dhalf]

This PR contains:

• Some changes/fixes to ML-related Python rules and/or their associated test file: lxml-in-pandas, numpy-in-pytorch-modules and pickles-in-pytorch
• A new rule to spot the usage of msgpack-numpy, a Python library that prepares numpy objects to be serialized with msgpack but that is backed by pickle. The rule excludes normal uses of msgpack (without the numpy translation layer).

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
2 out of 3 committers have signed the CLA.

✅ GrosQuildu
✅ dhalf
❌ Lucas Bourtoule

---

Lucas Bourtoule seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[GrosQuildu]

Thanks! Will review soon. Seems like you have to sign the CLA?

[dhalf]

Thanks! Will review soon. Seems like you have to sign the CLA?

Hi! I did sign twice but this doesn't seem to be reflected here... Any idea?

[dhalf]

I added two new rules to detect the use of Keras and Tensorflow loading functions. We reject all calls except those with a string literal (or constant) as file path. I also added a warning rule to deprecate the usage of pickle and hdf5 for Keras. The preferred format is now .keras which comes with checks that mitigate the risk of arbitrary code execution.

[GrosQuildu]

Overall looks good. Need @suhacker1 to review these (or somebody from ML). Then I can review from the semgrep/cq perspective.

[suhacker1]

From the ML-perspective, everything looks good to me!

[GrosQuildu]

Running python ./validate-metadata.py -s ./metadata-schema.yaml.schm -f . shows that technology and references metadata fields are missing

[mschwager]

LGTM! I left a few minor comments 👍