A Restricted Boltzmann machine with Bernoulli visible and hidden units. Parameters ---------- n_out : int The number of output dimensions/units. K : int The number of contrastive divergence steps to run before computing a
(self, n_out, K=1, init="glorot_uniform", optimizer=None)
| 364 | |
| 365 | class RBM(LayerBase): |
| 366 | def __init__(self, n_out, K=1, init="glorot_uniform", optimizer=None): |
| 367 | """ |
| 368 | A Restricted Boltzmann machine with Bernoulli visible and hidden units. |
| 369 | |
| 370 | Parameters |
| 371 | ---------- |
| 372 | n_out : int |
| 373 | The number of output dimensions/units. |
| 374 | K : int |
| 375 | The number of contrastive divergence steps to run before computing |
| 376 | a single gradient update. Default is 1. |
| 377 | init : {'glorot_normal', 'glorot_uniform', 'he_normal', 'he_uniform'} |
| 378 | The weight initialization strategy. Default is `'glorot_uniform'`. |
| 379 | optimizer : str, :doc:`Optimizer <numpy_ml.neural_nets.optimizers>` object, or None |
| 380 | The optimization strategy to use when performing gradient updates |
| 381 | within the :meth:`update` method. If None, use the :class:`SGD |
| 382 | <numpy_ml.neural_nets.optimizers.SGD>` optimizer with |
| 383 | default parameters. Default is None. |
| 384 | |
| 385 | Attributes |
| 386 | ---------- |
| 387 | X : list |
| 388 | Unused |
| 389 | gradients : dict |
| 390 | Dictionary of loss gradients with regard to the layer parameters |
| 391 | parameters : dict |
| 392 | Dictionary of layer parameters |
| 393 | hyperparameters : dict |
| 394 | Dictionary of layer hyperparameters |
| 395 | derived_variables : dict |
| 396 | Dictionary of any intermediate values computed during |
| 397 | forward/backward propagation. |
| 398 | """ # noqa: E501 |
| 399 | super().__init__(optimizer) |
| 400 | |
| 401 | self.K = K # CD-K |
| 402 | self.init = init |
| 403 | self.n_in = None |
| 404 | self.n_out = n_out |
| 405 | self.is_initialized = False |
| 406 | self.act_fn_V = ActivationInitializer("Sigmoid")() |
| 407 | self.act_fn_H = ActivationInitializer("Sigmoid")() |
| 408 | self.parameters = {"W": None, "b_in": None, "b_out": None} |
| 409 | |
| 410 | self._init_params() |
| 411 | |
| 412 | def _init_params(self): |
| 413 | init_weights = WeightInitializer(str(self.act_fn_V), mode=self.init) |