"Train" loss has declined but the test loss down to 0.6 or so began to rise

CodePudding user response:

Didn't quite understand the meaning of the original poster, usually only training dataset will have more than one batch and epoch, and the test dataset simply with input into the trained model to calculate a loss results, so I didn't understand what you said the test loss will also be changing

CodePudding user response:

Val - loss, training of wave in the test after the wave, there will be a acc and loss, the loss down to a certain degree is bigger

CodePudding user response:

May be a fitting, training too

CodePudding user response:

USES a lot of a fitting way to not make,

CodePudding user response:

Typically need to use a test data set, rather than repeatedly test, with multiple training model, training data set and control the better fitting, to do the test with test data set after the completion of training, the sample size of training data set don't too little,

CodePudding user response:

This is not very clear to me, everyone reads keras routine