ml_genn.metrics package

Metrics are used for calculating the performance of models based on some labels and the prediction obtained from a model using a ml_genn.readouts.Readout

class ml_genn.metrics.MeanSquareError

Bases: Metric

Computes the mean squared error between labels and prediction

reset()

Resets metric

property result

Quantity calculated by metric

update(y_true, y_pred, communicator)

Update metric based on a batch of true and predicted values.

Parameters:

  • y_true (ndarray) – ‘true’ values provided to compiled network evaluate/train method

  • y_pred (ndarray) – predicted values provided by model readout

  • communicator (Communicator | None) – communicator to use to synchronise metrics across GPUs when doing multi-GPU training.

class ml_genn.metrics.Metric

Bases: ABC

Base class for all metrics

abstract reset()

Resets metric

abstract property result: ndarray | None

Quantity calculated by metric

abstract update(y_true, y_pred, communicator)

Update metric based on a batch of true and predicted values.

Parameters:

  • y_true (ndarray) – ‘true’ values provided to compiled network evaluate/train method

  • y_pred (ndarray) – predicted values provided by model readout

  • communicator (Communicator | None) – communicator to use to synchronise metrics across GPUs when doing multi-GPU training.

class ml_genn.metrics.SparseCategoricalAccuracy

Bases: Metric

Computes the crossentropy between labels and prediction when there are two or more label classes, specified as integers.

reset()

Resets metric

property result

Quantity calculated by metric

update(y_true, y_pred, communicator)

Update metric based on a batch of true and predicted values.

Parameters:

  • y_true (ndarray) – ‘true’ values provided to compiled network evaluate/train method

  • y_pred (ndarray) – predicted values provided by model readout

  • communicator (Communicator | None) – communicator to use to synchronise metrics across GPUs when doing multi-GPU training.