Finding BIC value for fitted model

[yildizyasin]

There is an attribute returns AIC value for fitted model (FITTER.AIC_) but I couldn't find it for BIC. Is there a simple way to find BIC (Bayesian Information Criterion) value for fitter model?

exf = ExponentialFitter().fit(T, E, label="ExponentialFitter")
print(exf.AIC_) # works

print(exf.BIC_) # needed

[CamDavidsonPilon]

Hi @yildizyasin, it's not implemented. Not for any particular reason though. It would be a good addition.

[yildizyasin]

It is already implemented in find_best_method part. However, not sure if this calculation is correct. Or, am I missing something?
Here is a formula for BIC
https://machinelearningmastery.com/probabilistic-model-selection-measures/

The BIC statistic is calculated for logistic regression as follows (taken from “The Elements of Statistical Learning“):

BIC = -2 * LL + log(N) * k
Where log() has the base-e called the natural logarithm, LL is the log-likelihood of the model, N is the number of examples in the training dataset, and k is the number of parameters in the model.

I can implement it if we get aligned on this formula.

Now:

"BIC": lambda model: 2 * len(model._fitted_parameter_names)
- 2 * model.log_likelihood_ * np.log(model.event_observed.shape[0]),

Then:

"BIC_new": lambda model: -2 * model.log_likelihood_
        +  len(model._fitted_parameter_names) * np.log(model.event_observed.shape[0]),

[CamDavidsonPilon]

Yup, looks like I mixed up the terms. How silly! Can you write a PR to fix it?

[yildizyasin]

Just created pull request, #1275