Diagnostics / metrics

[BalzaniEdoardo]

This issue is a port from Alex's wish list for what it concerns the evaluation metrics

Diagnostics / metrics - (measures of performance specific to neural data)

• [x] poisson_log_likelihood
• [x] pseudo-r^2 McFadden
• [ ] Victor_pupura_dist (non-euclidian)
• [ ] Van_rossum_dist, kernel form neurips 2021 (Memming Arribas)
• [ ] outlier detection

For example, visualization tools to show example spikes that have very low log-likelihood

[billbrod]

As part of this, we should have a notebook describing the different metrics and why you would use one vs. the other, what context, etc. Currently glm.score includes a long discussion that belongs in this notebook