KeyError: 'Failed to format this callback filepath: "skintype_64_rmsprop_{val_loss:.3f}.h5". Reason: \'val_loss\''

Question

I have been trying to train my skin type classification model but it shows error.

model_name = f"skintype_{batch_size}_{optimizer}"
tensorboard = tf.keras.callbacks.TensorBoard(log_dir=os.path.join("logs", model_name))
# saves model checkpoint whenever we reach better weights
modelcheckpoint = tf.keras.callbacks.ModelCheckpoint(model_name + "_{val_loss:.3f}.h5", save_best_only=True, verbose=1)

history = m.fit(train_ds, validation_data=valid_ds, 
                steps_per_epoch=n_training_samples // batch_size, 
                validation_steps=n_validation_samples // batch_size, verbose=1, epochs=30,
                callbacks=[tensorboard, modelcheckpoint])

Answer 1

I don't know anything about Keras, and I only took a brief look at your code, but I think I see what's wrong. In the line:

modelcheckpoint = tf.keras.callbacks.ModelCheckpoint(model_name + "_{val_loss:.3f}.h5", save_best_only=True, verbose=1)

Keras is complaining about val_loss. I bet you want the format string containing that reference ("_{val_loss:.3f}.h5") to have a f at the front of it so that the reference gets interpreted and replaced with a value before it is passed into ModelCheckpoint. I think you need just a one character addition to get over this hurdle:

modelcheckpoint = tf.keras.callbacks.ModelCheckpoint(model_name + f"_{val_loss:.3f}.h5", save_best_only=True, verbose=1)

I hope I'm right and that this knowledge gets you cruising along again.

Answer 2

There could be a problem with the metrics you have mentioned in model.compile() as well.

For example, my ModelCheckpoint looks like this

path = "ops/model-{epoch:02d}-{val_categorical_accuracy:.2f}.hdf5"
checkpoint = ModelCheckpoint(filepath = path ,monitor ='val_categorical_accuracy',save_best_only=True,save_weights_only=True,mode='max',shuffle=True,verbose = 1,save_freq='epoch')

and my model.compile() looks like this:

model.compile(loss='binary_crossentropy',optimizer='adam',metrics=[tf.keras.metrics.CategoricalAccuracy()])

Most likely the reason why keras has an issue with val_loss is that you havent mentioned the metrics properly while compiling.

Answer 3

Have you checked the available storage on your Google Drive? If there is not enough storage, it cannot be saved.