jactorch.train.monitor#

Functions

`binary_classification_accuracy`(pred, label)	Compute the accuracy of binary classification.
`binary_classification_accuracy_4`(pred, label)
`classification_accuracy`(pred, label[, name])	Compute the accuracy of N-way classification.
`monitor_param_gradrms`(model)	Monitor the RMS of the gradients of the parameters of the given model.
`monitor_param_gradrms_ratio`(model)	Monitor the ratio of the RMS of the gradients of the parameters of the given model.
`monitor_param_rms`(model)	Monitor the RMS of the parameters of the given model.
`monitor_param_saturation`(model)	Monitor the saturation of the parameters of the given model.
`monitor_rms`(_dict, **values)	Monitor the RMS of the given values.
`regression_accuracy`(pred, label[, name])	Compute the accuracy of regression.

Functions

binary_classification_accuracy(pred, label, name='', saturation=True)[source]#

Compute the accuracy of binary classification.

Parameters:

pred (Tensor) – the prediction, of the same shape as label.
label (Tensor) – the label, of the same shape as pred.
name (str) – the name of this monitor.
saturation (bool) – whether to check the saturation of the prediction. Saturation is defined as \(1 - \min(pred, 1 - pred)\)

Returns:

a dict of monitor values.

Return type:

Dict[str, float]

binary_classification_accuracy_4(pred, label, name='')[source]#

Parameters:

pred (Tensor)
label (Tensor)
name (str)

Return type:

Dict[str, float]

classification_accuracy(pred, label, name='')[source]#

Compute the accuracy of N-way classification.

Parameters:

pred (Tensor) – the prediction, of the same shape as label.
label (Tensor) – the label, of the same shape as pred.
name (str) – the name of this monitor.

Returns:

a dict of monitor values.

Return type:

Dict[str, float]

monitor_param_gradrms(model)[source]#

Monitor the RMS of the gradients of the parameters of the given model.

Parameters:: model (Module) – the model to monitor.
Returns:: a dict of monitor values.
Return type:: Dict[str, float]

monitor_param_gradrms_ratio(model)[source]#

Monitor the ratio of the RMS of the gradients of the parameters of the given model.

Parameters:: model (Module) – the model to monitor.
Returns:: a dict of monitor values.
Return type:: Dict[str, float]

monitor_param_rms(model)[source]#

Monitor the RMS of the parameters of the given model.

Parameters:: model (Module) – the model to monitor.
Returns:: a dict of monitor values.
Return type:: Dict[str, float]

monitor_param_saturation(model)[source]#

Monitor the saturation of the parameters of the given model.

Parameters:: model (Module) – the model to monitor.
Returns:: a dict of monitor values.
Return type:: Dict[str, float]

monitor_rms(_dict, **values)[source]#

Monitor the RMS of the given values. This function takes either a dict or multiple keyword arguments.

Parameters:

_dict (Dict[str, Tensor] | None) – a dict of values.
**values (Tensor) – multiple keyword arguments.

Returns:

a dict of monitor values.

Return type:

Dict[str, float]

regression_accuracy(pred, label, name='')[source]#

Compute the accuracy of regression.

Parameters:

pred (Tensor) – the prediction, of the same shape as label.
label (Tensor) – the label, of the same shape as pred.
name (str) – the name of this monitor.

Returns:

a dict of monitor values.

Return type:

Dict[str, float]