Obtaining Raw Probabilities From OVR Logistic Regression

Question

I am relatively new to machine learning and I am working on a multi-class classification problem. I am attempting to utilize OVR logistic regression. When you run through an OVR model, the end result is a vector of standardized probabilities for each class that add up to 1. If I had 4 classes, my output might look like $[0.4, 0.2, 0.3, 0.1].$

I would like to obtain the raw probabilities for each class instead of the standardized probabilities. So I want what Binary Classifier 1 gives as the probability the observation belongs to class 1, what Binary Classifier 2 gives as the probability the observation belongs to class 2, etc.

If it helps, I want these probabilities because I am trying to introduce an intermediate step into my model where I create precision thresholds for each individual binary classifier. So for Binary Classifier 1 I want to line up the observations by their raw probability and find a threshold where the model is 80% precise and only make predictions on that subset of observations.

Any help is appreciated, thank you!

Answer 1

I can't speak to specific details about a package that does this, but you could simply run a logistic regression for each class. For each model you would code one class as positive and the rest as negative. In your example you would train 4 models.