Bug fix: monitor string must be val_accuracy

Without this fix, the model will not save, and the code will fail after the first pass of INCV when it tries to load a model that was not saved due to callback failure. This is a standard fix (see: tensorflow/tensorflow#33163).

INCV_main.py

@@ -158,7 +158,7 @@ def INCV_lr_schedule(epoch):
                                                   batch_size = batch_size),
                                   epochs = INCV_epochs,
                                   validation_data=(x_train[val1_idx,:], y_train_noisy[val1_idx,:]),
-                                  callbacks=[ModelCheckpoint(filepath=filepath_INCV, monitor='val_acc', verbose=1, save_best_only=INCV_save_best),
+                                  callbacks=[ModelCheckpoint(filepath=filepath_INCV, monitor='val_accuracy', verbose=1, save_best_only=INCV_save_best),
                                              noisy_acc,
                                              INCV_lr_callback])
     # Select samples of 'True' prediction