Mixing
Rewards normalising in reinforcement learning
I have 20 copies of an environment that gets a reward of 0.1 when it reaches it target and 0 otherwise.
What I want to do is make sense of how to normalise the rewards. Suppose I run the environment for 300 time steps. So the reward matrix is of size 300x20.
I usually normalise by doing:
discounted_rewards = torch.zeros_like(rewards, dtype=torch.float32, device=device)
for t in reversed(range(len(rewards))):
running_add = rewards[t] + discount * running_add
discounted_rewards[t] = running_add
mean = discounted_rewards.mean(0, keepdim=True)
std = discounted_rewards.std(0, keepdim=True) + 1e-10
discounted_rewards = (discounted_rewards - mean) / std
However, when I was doing an assignment I normalised the mean across the environments and not timesteps. i.e.
mean = discounted_rewards.mean(1, keepdim=True)
std = discounted_rewards.std(1, keepdim=True) + 1e-10
discounted_rewards = (discounted_rewards - mean) / std
and this seems to train faster. Was using PPO if it helps.
So my questions are:
- Which way are you supposed to normalise?
- I read this SO post which mentions that normalisation "doesn't mess with the sign of the gradient". However, if the reward is less than the mean it does change the sign of the gradient doesn't it?
- (optional) why does normalisation even work?
deep-learning reinforcement-learning
asked Nov 21 '18 at 0:44
sachinruksachinruk
2,12961933
add a comment |
I have 20 copies of an environment that gets a reward of 0.1 when it reaches it target and 0 otherwise.
What I want to do is make sense of how to normalise the rewards. Suppose I run the environment for 300 time steps. So the reward matrix is of size 300x20.
I usually normalise by doing:
discounted_rewards = torch.zeros_like(rewards, dtype=torch.float32, device=device)
for t in reversed(range(len(rewards))):
running_add = rewards[t] + discount * running_add
discounted_rewards[t] = running_add
mean = discounted_rewards.mean(0, keepdim=True)
std = discounted_rewards.std(0, keepdim=True) + 1e-10
discounted_rewards = (discounted_rewards - mean) / std
However, when I was doing an assignment I normalised the mean across the environments and not timesteps. i.e.
mean = discounted_rewards.mean(1, keepdim=True)
std = discounted_rewards.std(1, keepdim=True) + 1e-10
discounted_rewards = (discounted_rewards - mean) / std
and this seems to train faster. Was using PPO if it helps.
So my questions are:
- Which way are you supposed to normalise?
- I read this SO post which mentions that normalisation "doesn't mess with the sign of the gradient". However, if the reward is less than the mean it does change the sign of the gradient doesn't it?
- (optional) why does normalisation even work?
deep-learning reinforcement-learning
asked Nov 21 '18 at 0:44
sachinruksachinruk
2,12961933
add a comment |
I have 20 copies of an environment that gets a reward of 0.1 when it reaches it target and 0 otherwise.
What I want to do is make sense of how to normalise the rewards. Suppose I run the environment for 300 time steps. So the reward matrix is of size 300x20.
I usually normalise by doing:
discounted_rewards = torch.zeros_like(rewards, dtype=torch.float32, device=device)
for t in reversed(range(len(rewards))):
running_add = rewards[t] + discount * running_add
discounted_rewards[t] = running_add
mean = discounted_rewards.mean(0, keepdim=True)
std = discounted_rewards.std(0, keepdim=True) + 1e-10
discounted_rewards = (discounted_rewards - mean) / std
However, when I was doing an assignment I normalised the mean across the environments and not timesteps. i.e.
mean = discounted_rewards.mean(1, keepdim=True)
std = discounted_rewards.std(1, keepdim=True) + 1e-10
discounted_rewards = (discounted_rewards - mean) / std
and this seems to train faster. Was using PPO if it helps.
So my questions are:
- Which way are you supposed to normalise?
- I read this SO post which mentions that normalisation "doesn't mess with the sign of the gradient". However, if the reward is less than the mean it does change the sign of the gradient doesn't it?
- (optional) why does normalisation even work?
deep-learning reinforcement-learning
asked Nov 21 '18 at 0:44
sachinruksachinruk
2,12961933
I have 20 copies of an environment that gets a reward of 0.1 when it reaches it target and 0 otherwise.
What I want to do is make sense of how to normalise the rewards. Suppose I run the environment for 300 time steps. So the reward matrix is of size 300x20.
I usually normalise by doing:
discounted_rewards = torch.zeros_like(rewards, dtype=torch.float32, device=device)
for t in reversed(range(len(rewards))):
running_add = rewards[t] + discount * running_add
discounted_rewards[t] = running_add
mean = discounted_rewards.mean(0, keepdim=True)
std = discounted_rewards.std(0, keepdim=True) + 1e-10
discounted_rewards = (discounted_rewards - mean) / std
However, when I was doing an assignment I normalised the mean across the environments and not timesteps. i.e.
mean = discounted_rewards.mean(1, keepdim=True)
std = discounted_rewards.std(1, keepdim=True) + 1e-10
discounted_rewards = (discounted_rewards - mean) / std
and this seems to train faster. Was using PPO if it helps.
So my questions are:
- Which way are you supposed to normalise?
- I read this SO post which mentions that normalisation "doesn't mess with the sign of the gradient". However, if the reward is less than the mean it does change the sign of the gradient doesn't it?
- (optional) why does normalisation even work?
deep-learning reinforcement-learning
deep-learning reinforcement-learning
asked Nov 21 '18 at 0:44
sachinruksachinruk
2,12961933
asked Nov 21 '18 at 0:44
sachinruksachinruk
2,12961933
asked Nov 21 '18 at 0:44
sachinruksachinruk
2,12961933
asked Nov 21 '18 at 0:44
sachinruksachinruk
2,12961933
asked Nov 21 '18 at 0:44
sachinruksachinruk
2,12961933
2,12961933
add a comment |
add a comment |
0
active
oldest
votes
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%2f53403754%2frewards-normalising-in-reinforcement-learning%23new-answer', 'question_page');
}
);
Post as a guest
Required, but never shown
0
active
oldest
votes
0
active
oldest
votes
active
oldest
votes
active
oldest
votes
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%2f53403754%2frewards-normalising-in-reinforcement-learning%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
