Using nn.ModuleList over Python list dramatically slows down training
I'm training a very simple model that takes the number of hidden layers as a parameter. I originally stored these hidden layers in a vanilla python list , however when converting this list to a nn.ModuleList, training slows down dramatically by at least one order of magnitude!
AdderNet
class AdderNet(nn.Module):
def __init__(self, num_hidden, hidden_width):
super(AdderNet, self).__init__()
self.relu = nn.ReLU()
self.hiddenLayers =
self.inputLayer = nn.Linear(2, hidden_width)
self.outputLayer = nn.Linear(hidden_width, 1)
for i in range(num_hidden):
self.hiddenLayers.append(nn.Linear(hidden_width, hidden_width))
self.hiddenLayers = nn.ModuleList(self.hiddenLayers) # <--- causes DRAMATIC slowdown!
def forward(self, x):
out = self.inputLayer(x)
out = self.relu(out)
for layer in self.hiddenLayers:
out = layer(out)
out = self.relu(out)
return self.outputLayer(out)
Training
for epoch in range(num_epochs):
for i in range(0,len(data)):
out = model.forward(data[i].x)
loss = lossFunction(out, data[i].y)
optimizer.zero_grad()
loss.backward()
optimizer.step()
python neural-network pytorch
edited Nov 21 '18 at 5:27
Milo Lu
1,62711527
asked Nov 21 '18 at 0:52
Stephen LaskyStephen Lasky
140112
add a comment |
I'm training a very simple model that takes the number of hidden layers as a parameter. I originally stored these hidden layers in a vanilla python list , however when converting this list to a nn.ModuleList, training slows down dramatically by at least one order of magnitude!
AdderNet
class AdderNet(nn.Module):
def __init__(self, num_hidden, hidden_width):
super(AdderNet, self).__init__()
self.relu = nn.ReLU()
self.hiddenLayers =
self.inputLayer = nn.Linear(2, hidden_width)
self.outputLayer = nn.Linear(hidden_width, 1)
for i in range(num_hidden):
self.hiddenLayers.append(nn.Linear(hidden_width, hidden_width))
self.hiddenLayers = nn.ModuleList(self.hiddenLayers) # <--- causes DRAMATIC slowdown!
def forward(self, x):
out = self.inputLayer(x)
out = self.relu(out)
for layer in self.hiddenLayers:
out = layer(out)
out = self.relu(out)
return self.outputLayer(out)
Training
for epoch in range(num_epochs):
for i in range(0,len(data)):
out = model.forward(data[i].x)
loss = lossFunction(out, data[i].y)
optimizer.zero_grad()
loss.backward()
optimizer.step()
python neural-network pytorch
edited Nov 21 '18 at 5:27
Milo Lu
1,62711527
asked Nov 21 '18 at 0:52
Stephen LaskyStephen Lasky
140112
add a comment |
I'm training a very simple model that takes the number of hidden layers as a parameter. I originally stored these hidden layers in a vanilla python list , however when converting this list to a nn.ModuleList, training slows down dramatically by at least one order of magnitude!
AdderNet
class AdderNet(nn.Module):
def __init__(self, num_hidden, hidden_width):
super(AdderNet, self).__init__()
self.relu = nn.ReLU()
self.hiddenLayers =
self.inputLayer = nn.Linear(2, hidden_width)
self.outputLayer = nn.Linear(hidden_width, 1)
for i in range(num_hidden):
self.hiddenLayers.append(nn.Linear(hidden_width, hidden_width))
self.hiddenLayers = nn.ModuleList(self.hiddenLayers) # <--- causes DRAMATIC slowdown!
def forward(self, x):
out = self.inputLayer(x)
out = self.relu(out)
for layer in self.hiddenLayers:
out = layer(out)
out = self.relu(out)
return self.outputLayer(out)
Training
for epoch in range(num_epochs):
for i in range(0,len(data)):
out = model.forward(data[i].x)
loss = lossFunction(out, data[i].y)
optimizer.zero_grad()
loss.backward()
optimizer.step()
python neural-network pytorch
edited Nov 21 '18 at 5:27
Milo Lu
1,62711527
asked Nov 21 '18 at 0:52
Stephen LaskyStephen Lasky
140112
I'm training a very simple model that takes the number of hidden layers as a parameter. I originally stored these hidden layers in a vanilla python list , however when converting this list to a nn.ModuleList, training slows down dramatically by at least one order of magnitude!
AdderNet
class AdderNet(nn.Module):
def __init__(self, num_hidden, hidden_width):
super(AdderNet, self).__init__()
self.relu = nn.ReLU()
self.hiddenLayers =
self.inputLayer = nn.Linear(2, hidden_width)
self.outputLayer = nn.Linear(hidden_width, 1)
for i in range(num_hidden):
self.hiddenLayers.append(nn.Linear(hidden_width, hidden_width))
self.hiddenLayers = nn.ModuleList(self.hiddenLayers) # <--- causes DRAMATIC slowdown!
def forward(self, x):
out = self.inputLayer(x)
out = self.relu(out)
for layer in self.hiddenLayers:
out = layer(out)
out = self.relu(out)
return self.outputLayer(out)
Training
for epoch in range(num_epochs):
for i in range(0,len(data)):
out = model.forward(data[i].x)
loss = lossFunction(out, data[i].y)
optimizer.zero_grad()
loss.backward()
optimizer.step()
python neural-network pytorch
python neural-network pytorch
edited Nov 21 '18 at 5:27
Milo Lu
1,62711527
asked Nov 21 '18 at 0:52
Stephen LaskyStephen Lasky
140112
edited Nov 21 '18 at 5:27
Milo Lu
1,62711527
asked Nov 21 '18 at 0:52
Stephen LaskyStephen Lasky
140112
edited Nov 21 '18 at 5:27
Milo Lu
1,62711527
edited Nov 21 '18 at 5:27
Milo Lu
1,62711527
edited Nov 21 '18 at 5:27
Milo Lu
1,62711527
1,62711527
asked Nov 21 '18 at 0:52
Stephen LaskyStephen Lasky
140112
asked Nov 21 '18 at 0:52
Stephen LaskyStephen Lasky
140112
asked Nov 21 '18 at 0:52
Stephen LaskyStephen Lasky
140112
140112
add a comment |
add a comment |
1 Answer
1
active
oldest
votes
That's because when using a normal python list, the parameters are not added to the model's parameter list, but when using a ModuleList, they are. So, in the original scenario, you were never actually training the hidden layers, which is why it was faster. (Print out model.parameters() in each case and see what happens!)
answered Feb 19 at 19:56
user11086527user11086527
111
Really? So this means that during training that ONLY the input and output layer were being trained??
– Stephen Lasky
Feb 20 at 22:07
add a comment |
Your Answer
StackExchange.ifUsing("editor", function () {
StackExchange.using("externalEditor", function () {
StackExchange.using("snippets", function () {
StackExchange.snippets.init();
});
});
}, "code-snippets");
StackExchange.ready(function() {
var channelOptions = {
tags: "".split(" "),
id: "1"
};
initTagRenderer("".split(" "), "".split(" "), channelOptions);
StackExchange.using("externalEditor", function() {
// Have to fire editor after snippets, if snippets enabled
if (StackExchange.settings.snippets.snippetsEnabled) {
StackExchange.using("snippets", function() {
createEditor();
});
}
else {
createEditor();
}
});
function createEditor() {
StackExchange.prepareEditor({
heartbeatType: 'answer',
autoActivateHeartbeat: false,
convertImagesToLinks: true,
noModals: true,
showLowRepImageUploadWarning: true,
reputationToPostImages: 10,
bindNavPrevention: true,
postfix: "",
imageUploader: {
brandingHtml: "Powered by u003ca class="icon-imgur-white" href="https://imgur.com/"u003eu003c/au003e",
contentPolicyHtml: "User contributions licensed under u003ca href="https://creativecommons.org/licenses/by-sa/3.0/"u003ecc by-sa 3.0 with attribution requiredu003c/au003e u003ca href="https://stackoverflow.com/legal/content-policy"u003e(content policy)u003c/au003e",
allowUrls: true
},
onDemand: true,
discardSelector: ".discard-answer"
,immediatelyShowMarkdownHelp:true
});
}
});
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fstackoverflow.com%2fquestions%2f53403812%2fusing-nn-modulelist-over-python-list-dramatically-slows-down-training%23new-answer', 'question_page');
}
);
Post as a guest
Required, but never shown
1 Answer
1
active
oldest
votes
1 Answer
1
active
oldest
votes
active
oldest
votes
active
oldest
votes
That's because when using a normal python list, the parameters are not added to the model's parameter list, but when using a ModuleList, they are. So, in the original scenario, you were never actually training the hidden layers, which is why it was faster. (Print out model.parameters() in each case and see what happens!)
answered Feb 19 at 19:56
user11086527user11086527
111
Really? So this means that during training that ONLY the input and output layer were being trained??
– Stephen Lasky
Feb 20 at 22:07
add a comment |
That's because when using a normal python list, the parameters are not added to the model's parameter list, but when using a ModuleList, they are. So, in the original scenario, you were never actually training the hidden layers, which is why it was faster. (Print out model.parameters() in each case and see what happens!)
answered Feb 19 at 19:56
user11086527user11086527
111
Really? So this means that during training that ONLY the input and output layer were being trained??
– Stephen Lasky
Feb 20 at 22:07
add a comment |
That's because when using a normal python list, the parameters are not added to the model's parameter list, but when using a ModuleList, they are. So, in the original scenario, you were never actually training the hidden layers, which is why it was faster. (Print out model.parameters() in each case and see what happens!)
answered Feb 19 at 19:56
user11086527user11086527
111
That's because when using a normal python list, the parameters are not added to the model's parameter list, but when using a ModuleList, they are. So, in the original scenario, you were never actually training the hidden layers, which is why it was faster. (Print out model.parameters() in each case and see what happens!)
answered Feb 19 at 19:56
user11086527user11086527
111
answered Feb 19 at 19:56
user11086527user11086527
111
answered Feb 19 at 19:56
user11086527user11086527
111
answered Feb 19 at 19:56
user11086527user11086527
111
111
Really? So this means that during training that ONLY the input and output layer were being trained??
– Stephen Lasky
Feb 20 at 22:07
add a comment |
Really? So this means that during training that ONLY the input and output layer were being trained??
– Stephen Lasky
Feb 20 at 22:07
Really? So this means that during training that ONLY the input and output layer were being trained??
– Stephen Lasky
Feb 20 at 22:07
Really? So this means that during training that ONLY the input and output layer were being trained??
– Stephen Lasky
Feb 20 at 22:07
add a comment |
Thanks for contributing an answer to Stack Overflow!
- Please be sure to answer the question. Provide details and share your research!
But avoid …
- Asking for help, clarification, or responding to other answers.
- Making statements based on opinion; back them up with references or personal experience.
To learn more, see our tips on writing great answers.
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fstackoverflow.com%2fquestions%2f53403812%2fusing-nn-modulelist-over-python-list-dramatically-slows-down-training%23new-answer', 'question_page');
}
);
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
