evaluator module

Full Documentation for hippynn.experiment.evaluator module. Click here for a summary page.

Evaluating network performance over a dataset

class Evaluator(model, evaluation_loss, evaluation_loss_names, plot_maker=None, plot_every=1, db_info=None)

Bases: object

evaluate(generator, eval_type, when=None)

Construct metrics and activate plotter on a dataloader. Note: the model is performed on whatever device it resides on. The validation loss is expected to be on the CPU – this is not too slow, and allows for us to evaluate on large datasets whose results would take too much memory for the GPU. :param generator: the dataset loader :param eval_type: str describing dataset, such as ‘train’, ‘valid’, ‘test’ :param when: int (epoch #) or str describing when the evaluation is performed :return: loss_dict – dictionary mapping loss names to loss values for this generator

property var_list