salmon.triplets.samplers.adaptive.Embedding

class salmon.triplets.samplers.adaptive.Embedding(module, module__n: int = 85, module__d: int = 2, optimizer=<class 'torch.optim.adadelta.Adadelta'>, warm_start=True, max_epochs=100, initial_batch_size=512, **kwargs)

An optimization algorithm that produces an embedding from human responses of the form [head, winner, loser].

__init__(module, module__n: int = 85, module__d: int = 2, optimizer=<class 'torch.optim.adadelta.Adadelta'>, warm_start=True, max_epochs=100, initial_batch_size=512, **kwargs)

Parameters

modulenn.Module

The noise model to use.

module__nint

The number of items to embed.

module__dint, optional (default: 2)

The number of dimensions to embed into.

optimizertorch.optim.Optimizer

The optimizer to use.

max_epochsint, optional (default: 100)

The number of epochs—or passes through the dataset—to perform.

warm_startbool, optional (default: True)

Whether to use the existing embedding.

initial_batch_sizeint, optional (default: 512)

The optimizer’s (initial) batch size.

kwargsdict, optional

Additional keyword arguments to pass the underlying noise model (CKL, TSTE, etc).