probly.transformation.dropconnect¶

probly.transformation.dropconnect(base, p=0.25, rng_collection='dropconnect', rngs=1)[source]¶

Create a DropConnect predictor from a base predictor based on [MNM+19].

Parameters:

base (T) – The base model to be used for dropout.
p (float) – The probability of dropping out a neuron. Default is 0.25.
rng_collection (str) – Optional str for flax layer initialization. Default is “dropconnect”.
rngs (Rngs | RngStream | int) – Optional rngs for flax layer initialization (types: rnglib.Rngs | rnglib.RngStream | int), default: 1.

Returns:

The DropConnect predictor.

Return type: