What's the difference between capitalized and lowercase versions of Keras.losses and Keras.metrics?

Question

While I was constructing a simple Sequential ANN and selecting parameters for the model.compile method, I observed that Keras.metrics and Keras.losses contain capitalized as well as lowercase versions, for example tf.keras.metrics.SparseCategoricalAccuracy versus tf.keras.metrics.sparse_categorical_accuracy. I was wondering what the difference is between those versions and which one is more suitable to be used in model.compile ?

Answer 1

tf.keras.metrics.SparseCategoricalAccuracy is a Class so you get an object and you can pass it over to model.compile. Since it is an object it can have state between the calls.

However tf.keras.metrics.sparse_categorical_accuracy is a function and it is stateless. Both perform the same operation but their usage is different.