dcase_framework.keras_utils.StasherCallback

class dcase_framework.keras_utils.StasherCallback(*args, **kwargs)

Keras callback to monitor training process and store best model. Implements Keras Callback API.

This callback is very similar to standard ModelCheckpoint Keras callback, however it adds support for external metrics (metrics calculated outside Keras training process).

Constructor

Parameters:

epochs : int

Total amount of epochs

manual_update : bool

Manually update callback, use this to when injecting external metrics Default value True

monitor : str

Metric to monitor Default value ‘val_loss’

mode : str

Which way metric is interpreted, values {auto, min, max} Default value ‘auto’

period : int

Disable tqdm based progress bar Default value 1

initial_delay : int

Amount of epochs to wait at the beginning before quantity is monitored. Default value 10

save_weights : bool

Save weight to the disk Default value False

file_path : str

File name for model weight Default value None

__init__(*args, **kwargs)

Constructor

Parameters:

epochs : int

Total amount of epochs

manual_update : bool

Manually update callback, use this to when injecting external metrics Default value True

monitor : str

Metric to monitor Default value ‘val_loss’

mode : str

Which way metric is interpreted, values {auto, min, max} Default value ‘auto’

period : int

Disable tqdm based progress bar Default value 1

initial_delay : int

Amount of epochs to wait at the beginning before quantity is monitored. Default value 10

save_weights : bool

Save weight to the disk Default value False

file_path : str

File name for model weight Default value None

Methods

__init__(\*args, \*\*kwargs)	Constructor
add_external_metric(metric_label)
get_best()	Return best model seen
get_operator(metric)
log()	Print information about the best model into logging interface
on_batch_begin(batch[, logs])
on_batch_end(batch[, logs])
on_epoch_begin(epoch[, logs])
on_epoch_end(epoch[, logs])
on_train_begin([logs])
on_train_end([logs])
set_external_metric_value(metric_label, ...)	Add external metric value
set_model(model)
set_params(params)
update()