GPU porting (69c627e1) · Commits · 钟慕尧 / deepchem

contrib/torch/init.py→contrib/torch/torch_model.py

+4 −4

Original line number	Diff line number	Diff line
		@@ -159,16 +159,16 @@ class TorchMultitaskModel(Model):
		log("On batch %d" % ind, self.verbose)
		# Run training op.
		self.optimizer.zero_grad()
		X_b_input = torch.autograd.Variable(torch.FloatTensor(X_b))
		y_b_input = torch.autograd.Variable(torch.FloatTensor(y_b))
		w_b_input = torch.autograd.Variable(torch.FloatTensor(w_b))
		X_b_input = torch.autograd.Variable(torch.cuda.FloatTensor(X_b))
		y_b_input = torch.autograd.Variable(torch.cuda.FloatTensor(y_b))
		w_b_input = torch.autograd.Variable(torch.cuda.FloatTensor(w_b))
		outputs = self.forward(X_b_input, training=True)
		loss = self.add_training_cost(outputs, y_b_input, w_b_input)
		loss.backward()
		self.optimizer.step()
		avg_loss += loss
		n_batches += 1
		avg_loss = float(avg_loss.data.numpy()) / n_batches
		avg_loss = float(avg_loss.data.cpu().numpy()) / n_batches
		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)

contrib/torch/torch_multitask_classification.py

+9 −9

Original line number	Diff line number	Diff line
		@@ -8,7 +8,7 @@ Created on Mon Mar 13 22:31:24 2017
		import torch
		import numpy as np
		from deepchem.metrics import from_one_hot
		from deepchem.models.torch_models import TorchMultitaskModel
		from torch_model import TorchMultitaskModel


		class TorchMultitaskClassification(TorchMultitaskModel):
		@@ -62,10 +62,10 @@ class TorchMultitaskClassification(TorchMultitaskModel):
		for i in range(n_layers):
		W_init = np.random.normal(0, weight_init_stddevs[i],
		(prev_layer_size, layer_sizes[i]))
		W_init = torch.FloatTensor(W_init)
		W_init = torch.cuda.FloatTensor(W_init)
		self.W_list.append(torch.autograd.Variable(W_init, requires_grad=True))
		b_init = np.full((layer_sizes[i],), bias_init_consts[i])
		b_init = torch.FloatTensor(b_init)
		b_init = torch.cuda.FloatTensor(b_init)
		self.b_list.append(torch.autograd.Variable(b_init, requires_grad=True))
		prev_layer_size = layer_sizes[i]

		@@ -74,11 +74,11 @@ class TorchMultitaskClassification(TorchMultitaskModel):
		for i in range(self.n_tasks):
		W_init = np.random.normal(0, weight_init_stddevs[-1],
		(prev_layer_size, self.n_classes))
		W_init = torch.FloatTensor(W_init)
		W_init = torch.cuda.FloatTensor(W_init)
		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)
		b_init = torch.cuda.FloatTensor(b_init)
		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
		@@ -110,14 +110,14 @@ class TorchMultitaskClassification(TorchMultitaskModel):
		return loss

		def predict_on_batch(self, X_batch):
		X_batch = torch.autograd.Variable(torch.FloatTensor(X_batch))
		X_batch = torch.autograd.Variable(torch.cuda.FloatTensor(X_batch))
		outputs = self.forward(X_batch, training=False)
		y_pred_batch = torch.stack(outputs, 1).data.numpy()[:]
		y_pred_batch = torch.stack(outputs, 1).data.cpu().numpy()[:]
		y_pred_batch = from_one_hot(y_pred_batch, 2)
		return y_pred_batch

		def predict_proba_on_batch(self, X_batch):
		X_batch = torch.autograd.Variable(torch.FloatTensor(X_batch))
		X_batch = torch.autograd.Variable(torch.cuda.FloatTensor(X_batch))
		outputs = self.forward(X_batch, training=False)
		y_pred_batch = torch.stack(outputs, 1).data.numpy()[:]
		y_pred_batch = torch.stack(outputs, 1).data.cpu().numpy()[:]
		return y_pred_batch

contrib/torch/torch_multitask_regression.py

+7 −7

Original line number	Diff line number	Diff line
		@@ -7,7 +7,7 @@ Created on Mon Mar 13 22:31:24 2017

		import torch
		import numpy as np
		from deepchem.models.torch_models import TorchMultitaskModel
		from torch_model import TorchMultitaskModel


		class TorchMultitaskRegression(TorchMultitaskModel):
		@@ -60,10 +60,10 @@ class TorchMultitaskRegression(TorchMultitaskModel):
		for i in range(n_layers):
		W_init = np.random.normal(0, weight_init_stddevs[i],
		(prev_layer_size, layer_sizes[i]))
		W_init = torch.FloatTensor(W_init)
		W_init = torch.cuda.FloatTensor(W_init)
		self.W_list.append(torch.autograd.Variable(W_init, requires_grad=True))
		b_init = np.full((layer_sizes[i],), bias_init_consts[i])
		b_init = torch.FloatTensor(b_init)
		b_init = torch.cuda.FloatTensor(b_init)
		self.b_list.append(torch.autograd.Variable(b_init, requires_grad=True))
		prev_layer_size = layer_sizes[i]

		@@ -72,11 +72,11 @@ class TorchMultitaskRegression(TorchMultitaskModel):
		for i in range(self.n_tasks):
		W_init = np.random.normal(0, weight_init_stddevs[-1],
		(prev_layer_size, 1))
		W_init = torch.FloatTensor(W_init)
		W_init = torch.cuda.FloatTensor(W_init)
		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)
		b_init = torch.cuda.FloatTensor(b_init)
		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
		@@ -105,9 +105,9 @@ class TorchMultitaskRegression(TorchMultitaskModel):
		return loss

		def predict_on_batch(self, X_batch):
		X_batch = torch.autograd.Variable(torch.FloatTensor(X_batch))
		X_batch = torch.autograd.Variable(torch.cuda.FloatTensor(X_batch))
		outputs = self.forward(X_batch, training=False)
		y_pred_batch = torch.stack(outputs, 1).data.numpy()[:]
		y_pred_batch = torch.stack(outputs, 1).data.cpu().numpy()[:]
		y_pred_batch = np.squeeze(y_pred_batch, axis=2)
		return y_pred_batch

Admin message