Closed Catarina-Alves closed 3 years ago
After those two PRs have been accepted, I want to check these notebooks and update them if needed.
The example_spcc
notebook should be ignored in this PR. I modified here it by accident.
Hi @gnarayan, thank you for the feedback and useful comments!
I made a few modifications following the feedback:
pymultinest
in notebook 5, but this package should not be needed if all the algorithms are run with number_processes=1
. Were you forced to install it? If so, can you tell me which function triggered the behavior? pymultinest
is not installed through the major install because you can run the entire snmachine
pipeline without itValueError
issue in notebook 4 and no longer use the bw
input (this matplotlib input changed quite a bit recently)example_plasticc
are correct. They are much worse than the ones obtained through example_plasticc.ipynb
because the numbered notebooks are designed to be fast to run. Thus they use much fewer data and yield much worse results. I added a comment in notebook 7 highlighting this difference. Additionally, I highlighted at the beginning of example_plasticc.ipynb
that this notebook also uses a small dataset and therefore the results are not as good as the ones shown in the paper.Some of the GPs e.g. Objid 4220 look very overfit - some guidance on changing kernel parameters (or description of how they are chosen) could be useful.
Regarding the GPs, at the moment the 2D GP parameters are not tunable. I added a link to the section of the paper where they are described in notebook example_plasticc.ipynb
.
These notebooks show an example of how to use
snmachine
code.This PR can only be accepted after #240 and #243.