如何在 tensorflow 1.x 的每个训练时期保持模型的输出？

Question

I am trying to implement a constraint on the output of a neural network using the output of the previous training epoch. I tried using tf.assign() to update the value of a variable that holds the output, but it turned out that it holds the initial value.

Answer 1

You must use callbacks. It's my example for maximum scorу:

checkpoint_precision = ModelCheckpoint(filepath='best-weights, precision_selu_pr.hdf5', monitor='val_precision', mode='max', verbose=1, save_best_only=True)

checkpoint_auc = ModelCheckpoint(filepath='best-weights-auc_selu_pr.hdf5', monitor='val_auc', mode='max', verbose=1, save_best_only=True)

model.fit(x=data_x, y=Y, batch_size=100, epochs=100000, validation_data (x_val_scaled, Y_val), callbacks=[checkpoint_precision, checkpoint_auc])

For more information you can use this link: https://www.tensorflow.org/api_docs/python/tf/keras/callbacks