it seems to say on the Internet that accuracy in TF is calculated according to batch_size (denominator batch_size). Why is it that when batch_size = 4 is set, the output accuracy of each batch_size after updating the model parameters is not one of 0meme 0.25meme0.5mag0.75? But a result that is difficult to divide, such as 0.8763? I don"t know if there is any deviation in my understanding.