LossRecorder

A class for recording and retrieving losses for various terms across iterations during a training process.

Attributes

• iteration_losses (dict): Stores the total and individual term losses for each iteration.

• iteration_opt_parameters (dict): Keeps track of the optimization parameters used in each iteration.

• header_written (bool): Indicates if the header has been written to the output file.

Methods

• __init__(self): Initializes the LossRecorder class with necessary attributes.

• record_loss(self, iteration, opt_parameters, total_loss, each_term_loss, write_iteration_loss_filepath=None, note=None): Records the losses for a given iteration and optionally writes them to a specified file.

• get_iteration_loss(self, iteration): Retrieves the loss record for a specific iteration.

• get_losses(self): Returns a dictionary of all recorded losses for every iteration.

• get_term_loss(self, term): Retrieves a list of losses for a specified term across all recorded iterations.

• get_min_loss_and_iteration(self, term): Identifies the minimum loss value for a specified term across all iterations and the corresponding iteration number.

• write_losses_to_file(self, filepath): Writes the recorded losses to a file, formatting the output for readability.