Skip to content

[Feature] Make GAE return a 'value_target' entry #741

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Dec 13, 2022
Merged

[Feature] Make GAE return a 'value_target' entry #741

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
77e49fd
init
vmoens Dec 13, 2022
3854c81
docstr
vmoens Dec 13, 2022
d3ebfd2
nightly fix
vmoens Dec 13, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions torchrl/data/tensor_specs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -449,6 +449,11 @@ def _project(self, val: torch.Tensor) -> torch.Tensor:
maximum = maximum.expand_as(val)
val[val < minimum] = minimum[val < minimum]
val[val > maximum] = maximum[val > maximum]
except RuntimeError:
minimum = minimum.expand_as(val)
maximum = maximum.expand_as(val)
val[val < minimum] = minimum[val < minimum]
val[val > maximum] = maximum[val > maximum]
return val

def is_in(self, val: torch.Tensor) -> bool:
Expand Down
8 changes: 8 additions & 0 deletions torchrl/objectives/value/advantages.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -252,6 +252,13 @@ class GAE(nn.Module):
gradient_mode (bool): if True, gradients are propagated throught the computation of the value function.
Default is `False`.

GAE will return an :obj:`"advantage"` entry containing the advange value. It will also
return a :obj:`"value_target"` entry with the return value that is to be used
to train the value network. Finally, if :obj:`gradient_mode` is :obj:`True`,
an additional and differentiable :obj:`"value_error"` entry will be returned,
which simple represents the difference between the return and the value network
output (i.e. an additional distance loss should be applied to that signed value).

"""

def __init__(
Expand Down Expand Up @@ -336,6 +343,7 @@ def forward(
)

tensordict.set("advantage", adv.detach())
tensordict.set("value_target", value_target)
if self.gradient_mode:
tensordict.set("value_error", value_target - value)

Expand Down