Roc curves

[melmasri]

Hi Felix,

I updated the master branch to include

1. ROC plots. created a different folder for them in the benchmark director
2. fixed burning to prob for traj plots and autocorrelation.

One thing to look it is the error bard for the roc plots, I think your error bars consider FP/number of edges, while it should be (i think) FP/number of 0's.

Thanks

[melmasri]

Also the doc's in each folder do we change them manually. I need to update the brun in setup in this case.

[felixleopoldo]

1. Great. Once this is in place for MCMC modules, we could also have a dedicated module to produce ROC curves from the edge posterior probabilities.
2. Thanks. Yes, the docs has to be updated manually. Preferably the field description should be part of the JSON schema but in the meantime it is hard coded in docs.rst. The modules structure is described here and this part describes how to update the docs.

We use FP/P intentionally but FP/#0s is also interesting. Using TP/P and FP/P the SHD/P can be seen in the ROC-like plots as the Manhattan distance to the point (0,1). Have a look at page 31 in the Benchpress paper.

[melmasri]

🔨