Bias Initializer for unbiased classification #20357
Labels
keras-team-review-pending
Pending review by a Keras team member.
type:feature
The user is asking for a new feature.
Hi! I have been using the following initialization for the biases output layer of my classifiers, and it seems to work very well. In practice, I see it converge much faster (for the latest task I am using this for, it is taking 50 epochs on average instead of about 300, averaged across 10 holdouts). If you deem it useful, I can make a pull request to add it in.
Please do let me know whether I should do so.
The text was updated successfully, but these errors were encountered: