Open trevorlapay opened 6 years ago
Update: it didn't do anything good
Perhaps we could make a binary classifier that learns to distinguish between that problem class and all others.
Right, or figure out the problem words and eliminate them. The set minus the bad newsgroup classified at 87%, which is still pretty good...
On Sun, Sep 30, 2018, 12:30 PM Luke Hanks notifications@github.com wrote:
Perhaps we could make a binary classifier that learns to distinguish between that problem class and all others.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/trevorlapay/UNMFall18_ML_Project2_NB_LR/issues/3#issuecomment-425741081, or mute the thread https://github.com/notifications/unsubscribe-auth/ALDXeruZ97shf9U9qAGO9NhYldvhcV9kks5ugQ23gaJpZM4XAwkH .
I created a confusion matrix based on a split on the testing data and committed it to the repo. Interesting - index 7 newsgroup gets misclassified as every other type. I think one approach to getting better accuracy is to toss that class out so others don't get misclassified as that one. Running that now to see how we do.