Superluminous supernovae classifier for ZTF

[erusseil]

Brand new SLSN classifier for the ZTF stream.
It relies on feature extraction of light curve and the model has been trained on a high quality dataset with expert labels.
Currently slow and meaningless because it is meant to run after a drastic filter in the incomings alerts (trained almost only with transient) that should be added in the near future.
This first PR is a test to ensure that the module doesn't crash and isn't intended to run on the full stream.

[JulienPeloton]

Thanks @erusseil ! I've done a first pass on the code -- looks good to me. You can just update the CI script to enable the tests to run:

- name: Run test suites for ZTF
      run: |
        rm -f /tmp/forest_*.onnx
        pip install sncosmo==2.12.1  # Cette ligne là
        pip install xgboost==2.1.4   # Et cette ligne là
        ./run_tests.sh -s rubin
        curl -s https://codecov.io/bash | bash

Once the tests run, I will perform profiling on the code to see the bottlenecks, and we can think on pre-filtering to apply on the stream before this science module is applied.

[JulienPeloton]

Thanks @erusseil -- nearly there, I suggested some change to improve the performance :-)

[erusseil]

Thanks @JulienPeloton for the review ! I implemented all your requests :)
I will later work on improving the coverage of the doctests.

[JulienPeloton]

Thanks! So now with the current code, one has:

• score of -1.0 if not a valid transient
• score of 0.0 if a valid transient but not features extracted
• score > 0.0 if processed transients

Am I right?

[erusseil]

Yes you were right, but this is not an ideal behaviour in the end. I changed so code so that any invalid alert is reported as -1. That way 0 can directly be read as "the classifier gives a 0% chance of this events being a SLSN".