Add losses for VAE (3e7f114b) · Commits · 钟慕尧 / deepchem

deepchem/models/losses.py

+19 −4

Original line number	Diff line number	Diff line
		@@ -208,7 +208,7 @@ class SparseSoftmaxCrossEntropy(Loss):
		class VAE_ELBO(Loss):
		"""The Variational AutoEncoder loss, KL Divergence Regularize + marginal log-likelihood.

		This losses basesd on "Auto-Encoding Variational Bayes" (https://arxiv.org/abs/1312.6114).
		This losses based on _[1].
		ELBO(Evidence lower bound) lexically replaced Variational lower bound.
		BCE means marginal log-likelihood, and KLD means KL divergence with normal distribution.
		Added hyper parameter 'kl_scale' for KLD.
		@@ -239,6 +239,12 @@ class VAE_ELBO(Loss):
		Case pytorch
		>>> (VAE_ELBO()._create_pytorch_loss())(torch.tensor(logvar), torch.tensor(mu), torch.tensor(x), torch.tensor(reconstruction_x))
		tensor([0.7017, 0.7624], dtype=torch.float64)


		References
		----------
		.. [1] Diederik P Kingma., and Max Welling. "Auto-encoding variational bayes." arXiv preprint arXiv:1312.6114 (2013).

		"""

		def _compute_tf_loss(self, logvar, mu, x, reconstruction_x, kl_scale=1):
		@@ -266,7 +272,7 @@ class VAE_KLDivergence(Loss):
		"""The KL_divergence between hidden distribution and normal distribution.

		This loss represents KL divergence losses between normal distribution(using parameter of distribution)
		based on "Auto-Encoding Variational Bayes" (https://arxiv.org/abs/1312.6114).
		based on _[1].

		The logvar should have shape (batch_size, hidden_space) and each term represents
		standard deviation of hidden distribution. The mean shuold have
		@@ -291,6 +297,11 @@ class VAE_KLDivergence(Loss):
		Case pytorch
		>>> (VAE_KLDivergence()._create_pytorch_loss())(torch.tensor(logvar), torch.tensor(mu))
		tensor([0.1738, 0.5143], dtype=torch.float64)

		References
		----------
		.. [1] Diederik P Kingma., and Max Welling. "Auto-encoding variational bayes." arXiv preprint arXiv:1312.6114 (2013).

		"""

		def _compute_tf_loss(self, logvar, mu):
		@@ -316,8 +327,7 @@ class VAE_KLDivergence(Loss):
		class ShannonEntropy(Loss):
		"""The ShannonEntropy of discrete-distribution.

		This loss represents shannon entropy based on
		"A Brief Introduction to Shannon's Information Theory" (https://arxiv.org/abs/1612.09316).
		This loss represents shannon entropy based on _[1].

		The inputs should have shape (batch size, num of variable) and represents
		probabilites distribution.
		@@ -340,6 +350,11 @@ class ShannonEntropy(Loss):
		Case pytorch
		>>> (ShannonEntropy()._create_pytorch_loss())(torch.tensor(inputs))
		tensor([0.3054, 0.1625], dtype=torch.float64)

		References
		----------
		.. [1] Ricky Xiaofeng Chen. "A Brief Introduction to Shannon’s Information Theory." arXiv preprint arXiv:1612.09316 (2016).

		"""

		def _compute_tf_loss(self, inputs):

docs/models.rst

+9 −0

Original line number	Diff line number	Diff line
		@@ -199,6 +199,15 @@ Losses
		.. autoclass:: deepchem.models.losses.SparseSoftmaxCrossEntropy
		:members:

		.. autoclass:: deepchem.models.losses.VAE_ELBO
		:members:

		.. autoclass:: deepchem.models.losses.VAE_KLDivergence
		:members:

		.. autoclass:: deepchem.models.losses.ShannonEntropy
		:members:

		Optimizers
		----------

Admin message