Constrained Logistic Regression - How to apply specific constraints to logistic regression's coefficients

Last update: Dec 29, 2021

Overview

Constrained Logistic Regression

Sample implementation of constructing a logistic regression with given ranges on each of the feature's coefficients (via clogistic library).

The Data

We will use the processed version of telco customer churn data from Kaggle. The data can be downloaded here.

Steps

Define the constraints

For example:

# define constraints as dataframe
import numpy as np
constraint_df = pd.DataFrame(data=[
                                   ['gender',-np.inf,np.inf],
                                   ['SeniorCitizen',-np.inf,np.inf],
                                   ['Partner',-np.inf, 0],
                                   ['Dependents',-np.inf,0],
                                   ['tenure',-np.inf,0],
                                   ['PhoneService',-np.inf,0],
                                   ['PaperlessBilling',-np.inf,np.inf],
                                   ['MonthlyCharges',-np.inf,np.inf],
                                   ['intercept',-np.inf,np.inf]],
                             columns=['feature','lower_bound','upper_bound'])
constraint_df

|    | feature          |   lower_bound |   upper_bound |
|---:|:-----------------|--------------:|--------------:|
|  0 | gender           |          -inf |           inf |
|  1 | SeniorCitizen    |          -inf |           inf |
|  2 | Partner          |          -inf |             0 |
|  3 | Dependents       |          -inf |             0 |
|  4 | tenure           |          -inf |             0 |
|  5 | PhoneService     |          -inf |             0 |
|  6 | PaperlessBilling |          -inf |           inf |
|  7 | MonthlyCharges   |          -inf |           inf |
|  8 | intercept        |          -inf |           inf |

Model training via clogistic

# train using clogistic
from scipy.optimize import Bounds
from clogistic import LogisticRegression as clLogisticRegression

lower_bounds = constraint_df['lower_bound'].to_numpy()
upper_bounds = constraint_df['upper_bound'].to_numpy()
bounds = Bounds(lower_bounds, upper_bounds)

cl_logreg = clLogisticRegression(penalty='none')
cl_logreg.fit(X_train, y_train, bounds=bounds)

Retrieve the model coefficients

# coefficients as dataframe
cl_coef = pd.DataFrame({
    'feature': df.drop(columns='Churn').columns.tolist() + ['intercept'],
    'coefficient': list(cl_logreg.coef_[0]) + [cl_logreg.intercept_[0]]
})

cl_coef

|    | feature          |   coefficient |
|---:|:-----------------|--------------:|
|  0 | gender           |   0.0184168   |
|  1 | SeniorCitizen    |   0.506692    |
|  2 | Partner          |   3.85603e-09 |
|  3 | Dependents       |  -0.35721     |
|  4 | tenure           |  -0.0557211   |
|  5 | PhoneService     |  -0.796233    |
|  6 | PaperlessBilling |   0.398824    |
|  7 | MonthlyCharges   |   0.033197    |
|  8 | intercept        |  -1.36086     |

Constrained Logistic Regression - How to apply specific constraints to logistic regression's coefficients

Related tags

Overview

Constrained Logistic Regression

The Data

Steps

Define the constraints

Model training via clogistic

Owner

[NIPS 2021] UOTA: Improving Self-supervised Learning with Automated Unsupervised Outlier Arbitration.

Notspot robot simulation - Python version

A very simple baseline to estimate 2D & 3D SMPL-compatible keypoints from a single color image.

Revisiting, benchmarking, and refining Heterogeneous Graph Neural Networks.

Single-Shot Motion Completion with Transformer

3D-Transformer: Molecular Representation with Transformer in 3D Space

GitHub repository for the ICLR Computational Geometry & Topology Challenge 2021

MIMIC Code Repository: Code shared by the research community for the MIMIC-III database

A simple PyTorch Implementation of Generative Adversarial Networks, focusing on anime face drawing.

PyTorch Implementation of VAENAR-TTS: Variational Auto-Encoder based Non-AutoRegressive Text-to-Speech Synthesis.

Perturb-and-max-product: Sampling and learning in discrete energy-based models

This repository contains the official implementation code of the paper Improving Multimodal Fusion with Hierarchical Mutual Information Maximization for Multimodal Sentiment Analysis, accepted at EMNLP 2021.

Jittor is a high-performance deep learning framework based on JIT compiling and meta-operators.

clustering moroccan stocks time series data using k-means with dtw (dynamic time warping)

[ICCV2021] Learning to Track Objects from Unlabeled Videos

Official PyTorch implementation of PICCOLO: Point-Cloud Centric Omnidirectional Localization (ICCV 2021)

QTool: A Low-bit Quantization Toolbox for Deep Neural Networks in Computer Vision

Subnet Replacement Attack: Towards Practical Deployment-Stage Backdoor Attack on Deep Neural Networks

STARCH compuets regional extreme storm physical characteristics and moisture balance based on spatiotemporal precipitation data from reanalysis or climate model data.

The source code of the paper "Understanding Graph Neural Networks from Graph Signal Denoising Perspectives"