problem of 2021/assignment1/two_layer_net

cs231n / cs231n.github.io

Public facing notes page

MIT License

10.16k stars 4.06k forks source link

problem of 2021/assignment1/two_layer_net #277

Closed hustophi closed 2 years ago

hustophi commented 2 years ago

In two_layer_net.ipynb , when I use a Solver instance to train a TwoLayerNet, a numerical overflow error raised, which caused the softmax loss nan, but I have successfully passed all test above this step. Anyone also encountered this bug? Please give me some advice!

HumbleHumbert commented 1 year ago

same problem, I have seen many people just change the default learning rate to 1e-3, how did you fix it?