search

janfreyberg / superintendent

Practical active learning in python
https://superintendent.readthedocs.io
189 stars 18 forks source link

Models to use #63

Closed Lion-Mod closed 3 years ago

Lion-Mod commented 3 years ago

Jan love the idea of superintendent! I was thinking surely there is a way to efficiently label and in came active learning to save the day. Then in a jupyter environment.... love it!

My question is potentially a basic one but I would love to have an answer.

I'm looking to build a cnn image classifier. The data I have isn't labelled but I am willing to manually label some data hence superintendent.

Why are we using LogisticRegression? Reading the docs I understand this is not the be all end all but is good.

So would it be easy to hotswap in say a CNN or efficientnet into the ClassLabeller?

janfreyberg commented 3 years ago

Hi,

Yes definitely - it should be a model that's more suited than Logistic Regression. I only put that in because it's a fast demo, and for MNIST performs very well.

If you want to use any other model, simply make sure it conforms with the scikit learn fit / predict API. For example, keras or skorch neural nets do this.

Hope that makes sense!