Hi all, I'm implementing a simple sgd algorithm borrowing ideas from sklearn sgdregressor and vowpal wabbit many tricks. One feature I'm including is thecarlosjpita
In what scenario do you care about multiclass logistic loss? I thought this was always a proxy loss. -John On Tue, Mar 24, 2015 at 10:53 AM, Martin Popeljohn_c_langford
... No. VW currently cannot report [multi-class log loss](https://www.kaggle.com/wiki/MultiClassLogLoss). In `--oaa`, VW reports the 0/1 loss, i.e. 1-accuracy.
- Open Source
- Feb 15, 2011
- This is a public group.
- Attachments are permitted.
- Members cannot hide email address.
- Listed in Yahoo Groups directory.
- Membership does not require approval.
- Messages from new members require approval.
- Anyone can post messages.