notebook.community
Edit and run
squared hinge loss,有时也会用,在有些数据上表现好
negative one over number of classes
http://vision.stanford.edu/teaching/cs231n/linear-classify-demo/
选择正好能在内存中放下的batchsize,这里还要再听一下,1:00:00左右
没看懂
In [ ]: