when you planning to share code?

[Sandy4321]

when you planning to share code?

[adimajo]

2 separate issues there:

• updating (see ImportError: cannot import name 'generate_X_grid' #1 ) and porting all features of the R version of glmdisc (see adimajo/glmdisc)
• adding glmdisc-NN which is used in the ArXiV paper 1903.08920 (see adimajo/adimajo.github.io/assets/publications/GLMDISC_NN)

Tbh I have no idea... Before the end of the year I hope.

[adimajo]

An update:

• Still lacks interactions
• Ongoing work

[Sandy4321]

glmdisc: Discretization and Grouping for Logistic Regression
Can you clarify, how this code similar to R package?
My guess it is R code ported to Python?

[adimajo]

It is similar in the sense that both implement the glmdisc-SEM approach, i.e. a Stochastic-Expectation-Maximization algorithm to draw "good" discrete levels for continuous and factor levels (which yields a discretization / grouping function).

The R package can additionally search for interaction terms, i.e. features which "product" is included in the fitted logistic regression (say the combined occurrence of smoking and diabetes in predicting cancer), which this package has not (yet).

Parallel to that, another algorithm, glmdisc-NN, i.e. a neural network approach which uses neural networks to estimate "soft" discretization / grouping functions, is not yet properly implemented in either packages. It exists only in the form of Jupyter Notebooks (see adimajo/adimajo.github.io/assets/publications/GLMDISC_NN).

[Sandy4321]

Great can you share link to this Jupiter notebook?

[adimajo]

https://github.com/adimajo/adimajo.github.io/tree/master/assets/publications/GLMDISC_NN