sampepose / digit-recognizer

Classify MNIST digits
0 stars 0 forks source link

Logistic Regression Inaccurate #1

Open sampepose opened 8 years ago

sampepose commented 8 years ago

My implementation of softmax logistic classifier is ~70% accurate, whereas the scikit-learn classifier is ~90% accurate. I'm not sure why this is happening. I can get to ~80% accuracy if I remove the regularization.

dvfabbri commented 8 years ago

Hmm, lets talk about this tomorrow

On Tue, Feb 2, 2016 at 8:19 PM, sampepose notifications@github.com wrote:

My implementation of softmax logistic classifier is ~70% accurate, whereas the scikit-learn classifier is ~90% accurate. I'm not sure why this is happening. I can get to ~80% accuracy if I remove the regularization.

— Reply to this email directly or view it on GitHub https://github.com/sampepose/digit-recognizer/issues/1.