training loss
validation loss
As in the pictures, validation loss decreases normaly, training loss during one epoch is also decrease, but slowly. Why training loss decreases sharply between 2 epochs? How should I change parameters to map this gap?
I tried increasing batch_size, it works. But loss is even higher, precision and recall are worse. I made learning_rate smaller, but it dosen’t work. Someone says the model saw all data in epoch 1, so in epoch 2, it memorized the data, but I tried to decrease max_samples, loss is even higher. Is this big gap a problem? If yes, what parameter should I change?
Yang Cheng is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.