CODEWITHSUNDEEP
pythonscikit-learn
amarchin
amarchin

Reputation: 2114

How does sample_weight work in SGDClassifier?

I want to use importance sampling when I train the SGDClassifier. I've seen there is a sample_weight parameter in the fit and partial_fit methods but I am not sure how this parameter works.

Let's say that I have 10 samples and I use an array with 10 weights, will these weights multiply the loss function for the relative sample?

Upvotes: 2

Views: 532

Answers (1)

Grr
Grr

Reputation: 16079

You can find the relevant code in linear_model.sgd_fast, the most pertinent line being:

update *= class_weight * sample_weight

After each update step the final update is simply modified based on any provided sample or class weights provided.

An example of the high level result can be found in the user guide in SGD: Weighted samples

Upvotes: 3

Related Questions