Do not disable rascaline's logger

[frostedoyster]

Without these changes, evaluating the SOAP-BPNN looks like this on my laptop. However, I don't think this is a sustainable solution because we have to import rascaline

filippo@C03DL22:~/metatensor-models/tests/resources$ metatensor-models eval bpnn-model.pt eval.yaml
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
[2024-03-13 18:02:06][INFO] - Setting up evaluation set.
[2024-03-13 18:02:06][INFO] - Evaluating dataset
[2024-03-13 18:02:06][WARNING] - Forces not found in section 'energy'. Continue without forces!
[2024-03-13 18:02:06][WARNING] - Stress not found in section 'energy'. Continue without stress!
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
[2024-03-13 18:02:14][INFO] - energy RMSE: 0.27898045890441614
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)
rascaline::math::splines -- spline reached requested accuracy (1.000e-8) with 321 reference points (max absolute error is 5.699e-8)

---

📚 Documentation preview 📚: https://metatensor-models--147.org.readthedocs.build/en/147/

[Luthaf]

So this is problematic because it will also suppress the warnings. Maybe something like

from rascaline.log import RASCAL_LOG_LEVEL_WARN, RASCAL_LOG_LEVEL_ERROR

def rascaline_logging(level, message):
    if level in [RASCAL_LOG_LEVEL_WARN, RASCAL_LOG_LEVEL_ERROR]:
        rascaline.log.default_logging_callback(level, message)

rascaline.set_logging_callback(rascaline_logging)

Another solution would be to move the spline message to debug instead of info level in rascaline.

[frostedoyster]

Ok, fair, I will do this.

But how will we do this when rascaline won't be imported explicitly in Python?

[Luthaf]

You mean for eval?

Another solution would be to move the spline message to debug instead of info level in rascaline.

[frostedoyster]

Yes, I mean for eval

[frostedoyster]

Following Luthaf/rascaline#289, this pull request will be about removing the filtering of warnings and errors from rascaline

[Luthaf]

Sounds good, should we wait for the rascaline PR & also update the pinned commit?

[frostedoyster]

Yes, indeed