An optimizer that only performs a single pass over training data before updating model parameters.
Creates training batches out of the given cases.
Updates the parameters of the model based on the outputs computed during the forward pass.