Update multiclass_hinge_loss docstring

[kli-casia]

According to the code,
I think the original description

L_i = \\max_{j \\not = p_i} (0, t_j - t_{p_i} + \\delta)

Maybe should change to this

L_i = \\max_{j \\not = t_i} (0, p_j - p_{t_i} + \\delta)

where L_i is the loss for the i th sample, t_i is the ground truth label of the i th sample.
p is the predicted probability vector of the i th sample.

Becasue in the original description, t_{p_i} is meaningless since p_i is the predicted probabilty which is a decimal.

[f0k]

This makes the line too long. Please undo.

[f0k]

Can you squash this into a single commit? You'll need to clone the repository to your machine with git, then do git checkout patch-1, git reset --soft 9f10cad, git commit --amend --no-edit, and git push --force.

[f0k]

Will merge from the command line (to squash the two commits) after #846.

[Sentient07]

I'm not able to add @kli-nlpr 's fork of Lasagne to my remote. It shows unknown repository here (on the top). So I will open a new PR

[Sentient07]

I think this PR can be closed? (Master seems to be same as the diff that this proposes)

[f0k]

No, master still has the wrong version: https://github.com/Lasagne/Lasagne/blob/master/lasagne/objectives.py#L305
Otherwise this PR would have been marked as conflicting.
But I'll merge this myself, no need for a new PR, thanks!

[Sentient07]

No, master still has the wrong version

I had made this change previously and forgot to commit. It hence was moving around my other branches as well. My bad, sorry

[f0k]

Merged manually. Closing.