avalanche.evaluation.metrics.ForwardTransfer

class avalanche.evaluation.metrics.ForwardTransfer[source]

The standalone Forward Transfer metric. This metric returns the forward transfer relative to a specific key. Alternatively, this metric returns a dict in which each key is associated to the forward transfer. Forward transfer is computed as the difference between the value recorded for a specific key after the previous experience has been trained on, and random initialization before training. The value associated to a key can be updated with the update method.

At initialization, this metric returns an empty dictionary.

__init__()[source]

Creates an instance of the standalone Forward Transfer metric

Methods

__init__()

Creates an instance of the standalone Forward Transfer metric

reset()

Resets the metric internal state.

result([k])

param k:

the key for which returning forward transfer. If k is None,

update(k, v[, initial])

update_initial(k, v)

update_previous(k, v)

Attributes

initial

The initial value for each key.

previous

The previous experience value detected for each key