When calculating the loss, the params are vectors with same dimens, e.g. y_pred, y_true, and b. The calculation happens at the end of the epoch.

It means they did not pass 1 by 1 like [x_train1, y_train1, b1]. They first passing x_train, y_train to train the model each epoch, then calculate: y_pred, y_true with b in the custom loss.

def my_mse_loss_b(b):
def mseb(y_true, y_pred):
return K.mean(K.square(y_pred – y_true)) + b
return mseb

inputs = Input(shape=(200,))
x = Dense(128, activation=’relu’)(input)
x = Dense(200, activation=’relu’)(x)
outputs = Dense(200, activation=’linear’)(x)
model = Model(inputs=inputs, outputs=outputs)
model.compile(loss=my_mse_loss_b(B), optimizer=Adam(lr=0.0005))
history = model.fit(x_train, y_train, batch_size=64, epochs=2)

The shape of my x_train is (1000,200,1), y_train is (1000,200), B is (1000,)
When x_train[0] and y_train[0] pass to the training model, B[0] is passed as b, then x_train[1] and y_train[1] pass to the trainingmodel, B[1] is passed as b
Is that possible?
Would you please give me a hint?
Thanks, have a nice day!

You are welcome! 🙂