examples and yapfed (2f605d77) · Commits · 钟慕尧 / deepchem

contrib/torch/init.py

+11 −10

Original line number	Diff line number	Diff line
		@@ -12,7 +12,9 @@ from deepchem.trans import undo_transforms
		from deepchem.utils.save import log
		from deepchem.models import Model


		class TorchMultitaskModel(Model):

		def __init__(self,
		layer_sizes=[1000],
		weight_init_stddevs=[.02],
		@@ -106,7 +108,8 @@ class TorchMultitaskModel(Model):
		elif self.optimizer == 'adagrad':
		train_op = torch.optim.Adagrad(self.trainables, lr=self.learning_rate)
		elif self.optimizer == 'rmsprop':
		train_op = torch.optim.RMSprop(self.trainables, lr=self.learning_rate, momentum=self.momentum)
		train_op = torch.optim.RMSprop(
		self.trainables, lr=self.learning_rate, momentum=self.momentum)
		elif self.optimizer == 'sgd':
		train_op = torch.optim.SGD(self.trainables, lr=self.learning_rate)
		else:
		@@ -151,8 +154,7 @@ class TorchMultitaskModel(Model):
		# Turns out there are valid cases where we don't want pad-batches
		# on by default.
		#dataset.iterbatches(batch_size, pad_batches=True)):
		dataset.iterbatches(
		self.batch_size, pad_batches=self.pad_batches)):
		dataset.iterbatches(self.batch_size, pad_batches=self.pad_batches)):
		if ind % log_every_N_batches == 0:
		log("On batch %d" % ind, self.verbose)
		# Run training op.
		@@ -167,8 +169,7 @@ class TorchMultitaskModel(Model):
		avg_loss += loss
		n_batches += 1
		avg_loss = float(avg_loss.data.numpy()) / n_batches
		log('Ending epoch %d: Average loss %g' % (epoch, avg_loss),
		self.verbose)
		log('Ending epoch %d: Average loss %g' % (epoch, avg_loss), self.verbose)
		time2 = time.time()
		print("TIMING: model fitting took %0.3f s" % (time2 - time1), self.verbose)
		############################################################## TIMING

contrib/torch/torch_multitask_classification.py

+13 −11

Original line number	Diff line number	Diff line
		@@ -10,12 +10,10 @@ import numpy as np
		from deepchem.metrics import from_one_hot
		from deepchem.models.torch_models import TorchMultitaskModel


		class TorchMultitaskClassification(TorchMultitaskModel):
		def __init__(self,
		n_tasks,
		n_features,
		n_classes=2,
		**kwargs):

		def __init__(self, n_tasks, n_features, n_classes=2, **kwargs):
		"""Constructs the computational graph.

		This function constructs the computational graph for the model. It relies
		@@ -77,10 +75,12 @@ class TorchMultitaskClassification(TorchMultitaskModel):
		W_init = np.random.normal(0, weight_init_stddevs[-1],
		(prev_layer_size, self.n_classes))
		W_init = torch.FloatTensor(W_init)
		self.task_W_list.append(torch.autograd.Variable(W_init, requires_grad=True))
		self.task_W_list.append(
		torch.autograd.Variable(W_init, requires_grad=True))
		b_init = np.full((self.n_classes,), bias_init_consts[-1])
		b_init = torch.FloatTensor(b_init)
		self.task_b_list.append(torch.autograd.Variable(b_init, requires_grad=True))
		self.task_b_list.append(
		torch.autograd.Variable(b_init, requires_grad=True))
		self.trainables = self.W_list + self.b_list + self.task_W_list + self.task_b_list
		self.regularizaed_variables = self.W_list + self.task_W_list

		@@ -103,7 +103,9 @@ class TorchMultitaskClassification(TorchMultitaskModel):
		def cost(self, logit, label, weight):
		loss = []
		for i in range(logit.size()[0]):
		loss.append(torch.nn.functional.cross_entropy(logit[i,:], label[i].long()).mul(weight[i]))
		loss.append(
		torch.nn.functional.cross_entropy(logit[i, :], label[i].long()).mul(
		weight[i]))
		loss = torch.cat(loss).mean()
		return loss

contrib/torch/torch_multitask_regression.py

+10 −9

Original line number	Diff line number	Diff line
		@@ -9,11 +9,10 @@ import torch
		import numpy as np
		from deepchem.models.torch_models import TorchMultitaskModel


		class TorchMultitaskRegression(TorchMultitaskModel):
		def __init__(self,
		n_tasks,
		n_features,
		**kwargs):

		def __init__(self, n_tasks, n_features, **kwargs):
		"""Constructs the computational graph.

		This function constructs the computational graph for the model. It relies
		@@ -74,10 +73,12 @@ class TorchMultitaskRegression(TorchMultitaskModel):
		W_init = np.random.normal(0, weight_init_stddevs[-1],
		(prev_layer_size, 1))
		W_init = torch.FloatTensor(W_init)
		self.task_W_list.append(torch.autograd.Variable(W_init, requires_grad=True))
		self.task_W_list.append(
		torch.autograd.Variable(W_init, requires_grad=True))
		b_init = np.full((1,), bias_init_consts[-1])
		b_init = torch.FloatTensor(b_init)
		self.task_b_list.append(torch.autograd.Variable(b_init, requires_grad=True))
		self.task_b_list.append(
		torch.autograd.Variable(b_init, requires_grad=True))
		self.trainables = self.W_list + self.b_list + self.task_W_list + self.task_b_list
		self.regularizaed_variables = self.W_list + self.task_W_list

examples/delaney/delaney_torch.py

0 → 100644

+45 −0

Original line number	Diff line number	Diff line
		"""
		Script that trains multitask(torch) models on Delaney dataset.
		"""
		from __future__ import print_function
		from __future__ import division
		from __future__ import unicode_literals

		import sys
		sys.path.append('../../contrib/torch')
		from torch_multitask_regression import TorchMultitaskRegression
		import numpy as np
		import deepchem as dc

		# Only for debug!
		np.random.seed(123)

		# Load Delaney dataset
		n_features = 1024
		delaney_tasks, delaney_datasets, transformers = dc.molnet.load_delaney()
		train_dataset, valid_dataset, test_dataset = delaney_datasets

		# Fit models
		metric = dc.metrics.Metric(dc.metrics.pearson_r2_score, np.mean)

		model = TorchMultitaskRegression(
		len(delaney_tasks),
		n_features,
		layer_sizes=[1000],
		dropouts=[.25],
		learning_rate=0.001,
		batch_size=50,
		verbosity="high")

		# Fit trained model
		model.fit(train_dataset, nb_epoch=10)

		print("Evaluating model")
		train_scores = model.evaluate(train_dataset, [metric], transformers)
		valid_scores = model.evaluate(valid_dataset, [metric], transformers)

		print("Train scores")
		print(train_scores)

		print("Validation scores")
		print(valid_scores)

examples/tox21/tox21_torch.py

0 → 100644

+44 −0

Original line number	Diff line number	Diff line
		"""
		Script that trains multitask(torch) models on Tox21 dataset.
		"""
		from __future__ import print_function
		from __future__ import division
		from __future__ import unicode_literals

		import sys
		sys.path.append('../../contrib/torch')
		from torch_multitask_classification import TorchMultitaskClassification
		import numpy as np
		import deepchem as dc

		# Only for debug!
		np.random.seed(123)

		# Load Tox21 dataset
		n_features = 1024
		tox21_tasks, tox21_datasets, transformers = dc.molnet.load_tox21()
		train_dataset, valid_dataset, test_dataset = tox21_datasets

		# Fit models
		metric = dc.metrics.Metric(dc.metrics.roc_auc_score, np.mean)

		model = TorchMultitaskClassification(
		len(tox21_tasks),
		n_features,
		layer_sizes=[1000],
		dropouts=[.25],
		learning_rate=0.001,
		batch_size=50)

		# Fit trained model
		model.fit(train_dataset)

		print("Evaluating model")
		train_scores = model.evaluate(train_dataset, [metric], transformers)
		valid_scores = model.evaluate(valid_dataset, [metric], transformers)

		print("Train scores")
		print(train_scores)

		print("Validation scores")
		print(valid_scores)

Admin message