AmazaspShumik/sklearn-bayes

'BayesianLogisticRegression' object has no attribute '_mask_val'

Opened this issue · 7 comments

Deleted user commented

Hi,
Is this a known issue?
Cant even run fit ....
blr=BayesianLogisticRegression(n_iter=300, tol=1e-5, fit_intercept=False, verbose=True)

clf.fit(X_train, y_train)
File "/Library/Python/2.7/site-packages/skbayes/linear_models/bayes_logistic.py", line 87, in fit
y_bin[~mask] = self._mask_val
AttributeError: 'BayesianLogisticRegression' object has no attribute '_mask_val'

AmazaspShumik commented

BayesianLogisticRegression is superclass, fit method is available in EBLogisticRegression and VBLogisticRegression, which are two different versions of BayesianLogisticRegression

AmazaspShumik commented

VBLogisticRegresion uses Variational Bayes + Jaakola&Jordan
EBLogisticRegression uses Empirical Bayes + Laplace approximation.

Please use only algorithms listed here: https://github.com/AmazaspShumik/sklearn-bayes.

AmazaspShumik commented

I will add further comments to BayesianLogisticRegression class to avoid confusion

Deleted user commented

Hi,
Thanks :)

I was following the example here: https://github.com/AmazaspShumik/sklearn-bayes/blob/master/ipython_notebooks_tutorials/type_II_logistic/bayesian_logistic_demo.ipynb which clearly instantiates:
blr= BayesianLogisticRegression(tol_solver = 1e-5)

So maybe that should be fixed,
Regards,

AmazaspShumik commented

@QuantScientist

I am updating code for the whole package, and these are old tutorials.
I removed links to these old ipython notebooks from README but forgot to delete them (my mistake). I am working on new ipython notebooks for linear models, should update them next week.

Kind regards =)

AmazaspShumik commented

Hi,

I added several new ipython notebooks regarding classification:

  1. Logistic Regression with ARD prior (SBL algorithm) tutorial

  2. Variational Bayes Logistic Regression with ARD prior tutorial

  3. Bayesian Logistic Regression (includes both Empirical Bayes and Variational Bayes) tutorial

AmazaspShumik commented

@QuantScientist

I also deleted confusing tutorial, which used an old version of the package.
Hope this helps =)