monitor module

Monitor

A wrapper class for a neural network model which plots fairness metrics with the forward pass.

class monitor.Monitor(model, metrics_dict, groups)

Attributes:

• model the PyTorch Model Monitor is being used with
• metric_names list of the names of the metrics used
• metric_methods list of functions to evaluate metrics in the same order as metric_names
• groups list of the groups of data
• fig Matplotlib Pyplot figure with subplots of the value of each metric over training steps

get_fig()

Get fig attribute

Returns:Matplotlib Pyplot figure with subplots corresponding to each metric

update_fig(step_no, data, data_groups, labels)

Update figure at the end of a foward pass with values of metrics for each group and display the figure Assumes model outputs logits to be fed into softmax activation.

Parameters:

• step_no number of current training step (e.g. mini-batch number, batch number, epoch number)
• data validation data as list of inputs to model
• data_groups the group of each element in data as a 1D array
• labels labels for each element in data as a 1D array