Soft-Classifier Model

<aside> 💡 Check out the Google Colab implementation with both a binary and multi-class logistic classifier implemented from raw numpy.

</aside>

Implemented binary classifier.

Implemented multi-class classifier.

Overview

A Linear Logistic Classifier (aka Logistic Regression) is a soft-classifier model that can be binary (2-class) or easily extended to multi-class via changes in the loss function.

For binary classification, it provides valuable detail that hard classifiers omit:

• It is valuable to know if your model assigns 0.51 (threshold of 0.5) to a class vs. 0.9. For example, you may want to prompt the user for more detail if it falls within a certain margin of the threshold.
• This is also valuable from a loss calculation standpoint - if you are near the margin and you misclassify something, you are punished less than if the model assigns 0.99 to a value that turned out to be the other class.

For multi-class classifiers, the way you implement them can allow for similar interpretability. Further detail on each included below.

Implementation

In the linked Colab, I’ve built out a binary and multi-class logistic linear classifier from raw numpy. Let’s walk through the results:

Binary Classifier

Multi-Class Classifier