unexpected results

[321HG]

Why do I get unexpected results when I run the code as per the example provided?

[KindXiaoming]

that's very weird. well there's some code merge after the code was first released. This notebook should work at the time of first code release. If someone faces the same problem, please help. If looks weird to me because if it's random guessing, the acc would be around 0.5. The fact that train acc and test acc are both 0.0 is quite suspicious, looking like a bug. Could you print(correct, batch) in the acc function? or maybe it helps to make sure logit2 > logit1 and y[i] have the same dtype? Maybe some dtype conversion is needed (maybe this is a problem of numpy version?)