More than half the number of data set 0 category,
Training, the network of the predicted output, for the most part is 0,
But the model accuracy is higher, but is unusually high,
Other models and can't do that,

The losses and accuracy with tensorboard visualization, display losses probably down to 0.1,

Why is this PR curve?


Problem is my data?

CodePudding user response:

When training log is roughly such,

The output category labels,