Logistic Regression Package

Hi,
Has anybody released a package foe logistic and multinomial regressions?

We are planning to release one soon.
Couldnt find a direct package like that in sklearn.

Has anyone come across any such packages?

1 Like

Github search result

In https://github.com/mcreel/Econometrics there is code for general purpose maximum likelihood estimation. An example is the file https://github.com/mcreel/Econometrics/blob/master/Examples/MLE/EstimateLogit.jl, which shows how to estimate a logit model. Extending this to multinomial logit would be pretty easy. Here’s what the output of the example looks like:

julia> include("EstimateLogit.jl")
____________________________________________________________
estimate logit model
MLE Estimation Results    Convergence: true
Average Log-L: -0.67243   Observations: 30
Sandwich form covariance estimator

                estimate     st. err      t-stat     p-value
           1     0.40743     0.37948     1.07367     0.29214
           2    -0.07359     0.40602    -0.18126     0.85747

Information Criteria
                   Crit.      Crit/n
       CAIC     49.14795     1.63827
        BIC     47.14795     1.57160
        AIC     44.34556     1.47819
____________________________________________________________

I don’t have a package for it - but I have a super bare-bones logistic regression via SGD in my package.

https://github.com/caseykneale/ChemometricsTools.jl/blob/375619734d7c45659436ef861f2404c65bf42775/src/ClassificationModels.jl#L197-L201

I’d love to see a more fleshed out implementation.

https://juliastats.github.io/GLM.jl/stable/manual/ ?

2 Likes

Logistic and multinomial regressions are just models.

If you want to use these models for inference, there are quite a few choices to make (ML or Bayesian, or some other methodology like EM, various choices for regularization, validation eg PPC, …). Depending on these and the data size (1000 or 10^8 observations), or whether you are looking for a generic (GLM) setup, the complexity of implementations can range from simple to very complex.

Making your questions more specific would help.