Merge pull request #1567 from peastman/layers (c5a7c6ff) · Commits · 钟慕尧 / deepchem

deepchem/metalearning/maml.py

+3 −0

Original line number	Diff line number	Diff line
		@@ -101,6 +101,9 @@ class MAML(object):
		"""
		# Record inputs.

		raise Exception(
		'MAML does not currently work correctly. It needs to be rewritten to be compatible with modern TensorFlow'
		)
		self.learner = learner
		if isinstance(learner.loss, Layer):
		self._loss = learner.loss.out_tensor

deepchem/metalearning/tests/test_maml.py

+89 −90

Original line number	Diff line number	Diff line
		@@ -6,93 +6,92 @@ import numpy as np
		import tensorflow as tf
		import unittest


		class TestMAML(unittest.TestCase):

		@flaky
		def test_sine(self):
		"""Test meta-learning for sine function."""

		# This is a MetaLearner that learns to generate sine functions with variable
		# amplitude and phase.

		class SineLearner(dc.metalearning.MetaLearner):

		def __init__(self):
		self.batch_size = 10
		self.tg = dc.models.TensorGraph(use_queue=False)
		self.features = Feature(shape=(None, 1))
		self.labels = Label(shape=(None, 1))
		hidden1 = Dense(
		in_layers=self.features, out_channels=40, activation_fn=tf.nn.relu)
		hidden2 = Dense(
		in_layers=hidden1, out_channels=40, activation_fn=tf.nn.relu)
		output = Dense(in_layers=hidden2, out_channels=1)
		loss = L2Loss(in_layers=[output, self.labels])
		self.tg.add_output(output)
		self.tg.set_loss(loss)
		with self.tg._get_tf("Graph").as_default():
		self.tg.build()

		@property
		def loss(self):
		return self.tg.loss

		def select_task(self):
		self.amplitude = 5.0 * np.random.random()
		self.phase = np.pi * np.random.random()

		def get_batch(self):
		x = np.random.uniform(-5.0, 5.0, (self.batch_size, 1))
		feed_dict = {}
		feed_dict[self.features.out_tensor] = x
		feed_dict[self.labels.out_tensor] = self.amplitude * np.sin(
		x + self.phase)
		return feed_dict

		# Optimize it.

		learner = SineLearner()
		maml = dc.metalearning.MAML(learner, meta_batch_size=4)
		maml.fit(12000)

		# Test it out on some new tasks and see how it works.

		loss1 = []
		loss2 = []
		for i in range(50):
		learner.select_task()
		feed_dict = learner.get_batch()
		for key, value in learner.get_batch().items():
		feed_dict[maml._meta_placeholders[key]] = value
		loss1.append(
		np.average(
		np.sqrt(maml._session.run(maml._loss, feed_dict=feed_dict))))
		loss2.append(
		np.average(
		np.sqrt(maml._session.run(maml._meta_loss, feed_dict=feed_dict))))

		# Initially the model should do a bad job of fitting the sine function.

		assert np.average(loss1) > 1.0

		# After one step of optimization it should do much better.

		assert np.average(loss2) < 1.0

		# Verify that we can create a new MAML object, reload the parameters from the first one, and
		# get the same result.

		new_maml = dc.metalearning.MAML(learner, model_dir=maml.model_dir)
		new_maml.restore()
		new_loss = np.average(
		np.sqrt(new_maml._session.run(new_maml._loss, feed_dict=feed_dict)))
		assert new_loss == loss1[-1]

		# Do the same thing, only using the "restore" argument to fit().

		new_maml = dc.metalearning.MAML(learner, model_dir=maml.model_dir)
		new_maml.fit(0, restore=True)
		new_loss = np.average(
		np.sqrt(new_maml._session.run(new_maml._loss, feed_dict=feed_dict)))
		assert new_loss == loss1[-1]
		# class TestMAML(unittest.TestCase):
		#
		# @flaky
		# def test_sine(self):
		# """Test meta-learning for sine function."""
		#
		# # This is a MetaLearner that learns to generate sine functions with variable
		# # amplitude and phase.
		#
		# class SineLearner(dc.metalearning.MetaLearner):
		#
		# def __init__(self):
		# self.batch_size = 10
		# self.tg = dc.models.TensorGraph(use_queue=False)
		# self.features = Feature(shape=(None, 1))
		# self.labels = Label(shape=(None, 1))
		# hidden1 = Dense(
		# in_layers=self.features, out_channels=40, activation_fn=tf.nn.relu)
		# hidden2 = Dense(
		# in_layers=hidden1, out_channels=40, activation_fn=tf.nn.relu)
		# output = Dense(in_layers=hidden2, out_channels=1)
		# loss = L2Loss(in_layers=[output, self.labels])
		# self.tg.add_output(output)
		# self.tg.set_loss(loss)
		# with self.tg._get_tf("Graph").as_default():
		# self.tg.build()
		#
		# @property
		# def loss(self):
		# return self.tg.loss
		#
		# def select_task(self):
		# self.amplitude = 5.0 * np.random.random()
		# self.phase = np.pi * np.random.random()
		#
		# def get_batch(self):
		# x = np.random.uniform(-5.0, 5.0, (self.batch_size, 1))
		# feed_dict = {}
		# feed_dict[self.features.out_tensor] = x
		# feed_dict[self.labels.out_tensor] = self.amplitude * np.sin(
		# x + self.phase)
		# return feed_dict
		#
		# # Optimize it.
		#
		# learner = SineLearner()
		# maml = dc.metalearning.MAML(learner, meta_batch_size=4)
		# maml.fit(12000)
		#
		# # Test it out on some new tasks and see how it works.
		#
		# loss1 = []
		# loss2 = []
		# for i in range(50):
		# learner.select_task()
		# feed_dict = learner.get_batch()
		# for key, value in learner.get_batch().items():
		# feed_dict[maml._meta_placeholders[key]] = value
		# loss1.append(
		# np.average(
		# np.sqrt(maml._session.run(maml._loss, feed_dict=feed_dict))))
		# loss2.append(
		# np.average(
		# np.sqrt(maml._session.run(maml._meta_loss, feed_dict=feed_dict))))
		#
		# # Initially the model should do a bad job of fitting the sine function.
		#
		# assert np.average(loss1) > 1.0
		#
		# # After one step of optimization it should do much better.
		#
		# assert np.average(loss2) < 1.0
		#
		# # Verify that we can create a new MAML object, reload the parameters from the first one, and
		# # get the same result.
		#
		# new_maml = dc.metalearning.MAML(learner, model_dir=maml.model_dir)
		# new_maml.restore()
		# new_loss = np.average(
		# np.sqrt(new_maml._session.run(new_maml._loss, feed_dict=feed_dict)))
		# assert new_loss == loss1[-1]
		#
		# # Do the same thing, only using the "restore" argument to fit().
		#
		# new_maml = dc.metalearning.MAML(learner, model_dir=maml.model_dir)
		# new_maml.fit(0, restore=True)
		# new_loss = np.average(
		# np.sqrt(new_maml._session.run(new_maml._loss, feed_dict=feed_dict)))
		# assert new_loss == loss1[-1]

deepchem/models/tensorgraph/IRV.py

+4 −4

Original line number	Diff line number	Diff line
		@@ -64,7 +64,7 @@ class IRVLayer(Layer):
		out_tensor = tf.concat(outputs, axis=1)

		if set_tensors:
		self.variables = self.trainable_weights
		self.trainable_variables = self.trainable_weights
		self.out_tensor = out_tensor
		return out_tensor

		@@ -72,12 +72,12 @@ class IRVLayer(Layer):
		V, W, b, b2 = self.V, self.W, self.b, self.b2
		self.V, self.W, self.b, self.b2 = None, None, None, None

		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.variables
		self.out_tensor, self.trainable_weights, self.variables = None, [], []
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.trainable_variables
		self.out_tensor, self.trainable_weights, self.trainable_variables = None, [], []
		return V, W, b, b2, out_tensor, trainable_weights, variables

		def set_tensors(self, tensor):
		self.V, self.W, self.b, self.b2, self.out_tensor, self.trainable_weights, self.variables = tensor
		self.V, self.W, self.b, self.b2, self.out_tensor, self.trainable_weights, self.trainable_variables = tensor


		class IRVRegularize(Layer):

deepchem/models/tensorgraph/graph_layers.py

+36 −36

Original line number	Diff line number	Diff line
		@@ -166,7 +166,7 @@ class WeaveLayer(Layer):

		self.out_tensors = [A, P]
		if set_tensors:
		self.variables = self.trainable_weights
		self.trainable_variables = self.trainable_weights
		self.out_tensor = A
		return self.out_tensors

		@@ -177,8 +177,8 @@ class WeaveLayer(Layer):
		W_AA, b_AA, W_PA, b_PA, W_A, b_A = self.W_AA, self.b_AA, self.W_PA, self.b_PA, self.W_A, self.b_A
		self.W_AA, self.b_AA, self.W_PA, self.b_PA, self.W_A, self.b_A = None, None, None, None, None, None

		out_tensor, out_tensors, trainable_weights, variables = self.out_tensor, self.out_tensors, self.trainable_weights, self.variables
		self.out_tensor, self.out_tensors, self.trainable_weights, self.variables = None, [], [], []
		out_tensor, out_tensors, trainable_weights, variables = self.out_tensor, self.out_tensors, self.trainable_weights, self.trainable_variables
		self.out_tensor, self.out_tensors, self.trainable_weights, self.trainable_variables = None, [], [], []

		return W_AP, b_AP, W_PP, W_PP, W_P, b_P, \
		W_AA, b_AA, W_PA, b_PA, W_A, b_A, \
		@@ -187,7 +187,7 @@ class WeaveLayer(Layer):
		def set_tensors(self, tensor):
		self.W_AP, self.b_AP, self.W_PP, self.W_PP, self.W_P, self.b_P, \
		self.W_AA, self.b_AA, self.W_PA, self.b_PA, self.W_A, self.b_A, \
		self.out_tensor, self.out_tensors, self.trainable_weights, self.variables = tensor
		self.out_tensor, self.out_tensors, self.trainable_weights, self.trainable_variables = tensor


		def WeaveLayerFactory(**kwargs):
		@@ -266,7 +266,7 @@ class WeaveGather(Layer):

		out_tensor = output_molecules
		if set_tensors:
		self.variables = self.trainable_weights
		self.trainable_variables = self.trainable_weights
		self.out_tensor = out_tensor
		return out_tensor

		@@ -290,12 +290,12 @@ class WeaveGather(Layer):
		W, b = self.W, self.b
		self.W, self.b = None, None

		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.variables
		self.out_tensor, self.trainable_weights, self.variables = None, [], []
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.trainable_variables
		self.out_tensor, self.trainable_weights, self.trainable_variables = None, [], []
		return W, b, out_tensor, trainable_weights, variables

		def set_tensors(self, tensor):
		self.W, self.b, self.out_tensor, self.trainable_weights, self.variables = tensor
		self.W, self.b, self.out_tensor, self.trainable_weights, self.trainable_variables = tensor


		class DTNNEmbedding(Layer):
		@@ -341,19 +341,19 @@ class DTNNEmbedding(Layer):
		atom_features = tf.nn.embedding_lookup(self.embedding_list, atom_number)
		out_tensor = atom_features
		if set_tensors:
		self.variables = self.trainable_weights
		self.trainable_variables = self.trainable_weights
		self.out_tensor = atom_features
		return out_tensor

		def none_tensors(self):
		embedding_list = self.embedding_list
		self.embedding_list = None
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.variables
		self.out_tensor, self.trainable_weights, self.variables = None, [], []
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.trainable_variables
		self.out_tensor, self.trainable_weights, self.trainable_variables = None, [], []
		return embedding_list, out_tensor, trainable_weights, variables

		def set_tensors(self, tensor):
		self.embedding_list, self.out_tensor, self.trainable_weights, self.variables = tensor
		self.embedding_list, self.out_tensor, self.trainable_weights, self.trainable_variables = tensor


		class DTNNStep(Layer):
		@@ -436,19 +436,19 @@ class DTNNStep(Layer):
		distance_membership_i) - output_ii + atom_features
		out_tensor = outputs
		if set_tensors:
		self.variables = self.trainable_weights
		self.trainable_variables = self.trainable_weights
		self.out_tensor = out_tensor
		return out_tensor

		def none_tensors(self):
		W_cf, W_df, W_fc, b_cf, b_df = self.W_cf, self.W_df, self.W_fc, self.b_cf, self.b_df
		self.W_cf, self.W_df, self.W_fc, self.b_cf, self.b_df = None, None, None, None, None
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.variables
		self.out_tensor, self.trainable_weights, self.variables = None, [], []
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.trainable_variables
		self.out_tensor, self.trainable_weights, self.trainable_variables = None, [], []
		return W_cf, W_df, W_fc, b_cf, b_df, out_tensor, trainable_weights, variables

		def set_tensors(self, tensor):
		self.W_cf, self.W_df, self.W_fc, self.b_cf, self.b_df, self.out_tensor, self.trainable_weights, self.variables = tensor
		self.W_cf, self.W_df, self.W_fc, self.b_cf, self.b_df, self.out_tensor, self.trainable_weights, self.trainable_variables = tensor


		class DTNNGather(Layer):
		@@ -524,19 +524,19 @@ class DTNNGather(Layer):
		output = tf.segment_sum(output, atom_membership)
		out_tensor = output
		if set_tensors:
		self.variables = self.trainable_weights
		self.trainable_variables = self.trainable_weights
		self.out_tensor = out_tensor
		return out_tensor

		def none_tensors(self):
		W_list, b_list = self.W_list, self.b_list
		self.W_list, self.b_list = [], []
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.variables
		self.out_tensor, self.trainable_weights, self.variables = None, [], []
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.trainable_variables
		self.out_tensor, self.trainable_weights, self.trainable_variables = None, [], []
		return W_list, b_list, out_tensor, trainable_weights, variables

		def set_tensors(self, tensor):
		self.W_list, self.b_list, self.out_tensor, self.trainable_weights, self.variables = tensor
		self.W_list, self.b_list, self.out_tensor, self.trainable_weights, self.trainable_variables = tensor


		class DTNNExtract(Layer):
		@@ -694,7 +694,7 @@ class DAGLayer(Layer):

		out_tensor = batch_outputs
		if set_tensors:
		self.variables = self.trainable_weights
		self.trainable_variables = self.trainable_weights
		self.out_tensor = out_tensor
		return out_tensor

		@@ -711,12 +711,12 @@ class DAGLayer(Layer):
		def none_tensors(self):
		W_list, b_list = self.W_list, self.b_list
		self.W_list, self.b_list = [], []
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.variables
		self.out_tensor, self.trainable_weights, self.variables = None, [], []
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.trainable_variables
		self.out_tensor, self.trainable_weights, self.trainable_variables = None, [], []
		return W_list, b_list, out_tensor, trainable_weights, variables

		def set_tensors(self, tensor):
		self.W_list, self.b_list, self.out_tensor, self.trainable_weights, self.variables = tensor
		self.W_list, self.b_list, self.out_tensor, self.trainable_weights, self.trainable_variables = tensor


		class DAGGather(Layer):
		@@ -803,7 +803,7 @@ class DAGGather(Layer):
		**kwargs)
		out_tensor = outputs
		if set_tensors:
		self.variables = self.trainable_weights
		self.trainable_variables = self.trainable_weights
		self.out_tensor = out_tensor
		return out_tensor

		@@ -820,12 +820,12 @@ class DAGGather(Layer):
		def none_tensors(self):
		W_list, b_list = self.W_list, self.b_list
		self.W_list, self.b_list = [], []
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.variables
		self.out_tensor, self.trainable_weights, self.variables = None, [], []
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.trainable_variables
		self.out_tensor, self.trainable_weights, self.trainable_variables = None, [], []
		return W_list, b_list, out_tensor, trainable_weights, variables

		def set_tensors(self, tensor):
		self.W_list, self.b_list, self.out_tensor, self.trainable_weights, self.variables = tensor
		self.W_list, self.b_list, self.out_tensor, self.trainable_weights, self.trainable_variables = tensor


		class MessagePassing(Layer):
		@@ -898,19 +898,19 @@ class MessagePassing(Layer):
		out_tensor = out

		if set_tensors:
		self.variables = self.trainable_weights
		self.trainable_variables = self.trainable_weights
		self.out_tensor = out_tensor
		return out_tensor

		def none_tensors(self):
		message_tensors = self.message_function.none_tensors()
		update_tensors = self.update_function.none_tensors()
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.variables
		self.out_tensor, self.trainable_weights, self.variables = None, [], []
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.trainable_variables
		self.out_tensor, self.trainable_weights, self.trainable_variables = None, [], []
		return message_tensors, update_tensors, out_tensor, trainable_weights, variables

		def set_tensors(self, tensor):
		message_tensors, update_tensors, self.out_tensor, self.trainable_weights, self.variables = tensor
		message_tensors, update_tensors, self.out_tensor, self.trainable_weights, self.trainable_variables = tensor
		self.message_function.set_tensors(message_tensors)
		self.update_function.set_tensors(update_tensors)

		@@ -1048,7 +1048,7 @@ class SetGather(Layer):

		out_tensor = q_star
		if set_tensors:
		self.variables = self.trainable_weights
		self.trainable_variables = self.trainable_weights
		self.out_tensor = out_tensor
		return out_tensor

		@@ -1067,9 +1067,9 @@ class SetGather(Layer):
		def none_tensors(self):
		U, b, c, h = self.U, self.b, self.c, self.h
		self.U, self.b, self.c, self.h = None, None, None, None
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.variables
		self.out_tensor, self.trainable_weights, self.variables = None, [], []
		out_tensor, trainable_weights, variables = self.out_tensor, self.trainable_weights, self.trainable_variables
		self.out_tensor, self.trainable_weights, self.trainable_variables = None, [], []
		return U, b, c, h, out_tensor, trainable_weights, variables

		def set_tensors(self, tensor):
		self.U, self.b, self.c, self.h, self.out_tensor, self.trainable_weights, self.variables = tensor
		self.U, self.b, self.c, self.h, self.out_tensor, self.trainable_weights, self.trainable_variables = tensor

deepchem/models/tensorgraph/layers.py

+124 −234

File changed.

Preview size limit exceeded, changes collapsed.

Admin message