Skip to content
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

[Feature] ActionDiscretizer custom sampling #2609

Merged
merged 1 commit into from
Dec 3, 2024
Merged

[Feature] ActionDiscretizer custom sampling #2609

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
46 changes: 24 additions & 22 deletions torchrl/envs/transforms/transforms.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8583,6 +8583,26 @@ def _indent(s):
f"\n{_indent(out_action_key)},\n{_indent(sampling)},\n{_indent(categorical)})"
)

def _custom_arange(self, nint, device):
result = torch.arange(
start=0.0,
end=1.0,
step=1 / nint,
dtype=self.dtype,
device=device,
)
result_ = result
if self.sampling in (
self.SamplingStrategy.HIGH,
self.SamplingStrategy.MEDIAN,
):
result_ = (1 - result).flip(0)
if self.sampling == self.SamplingStrategy.MEDIAN:
result = (result + result_) / 2
else:
result = result_
return result

def transform_input_spec(self, input_spec):
try:
action_spec = self.parent.full_action_spec_unbatched[self.in_keys_inv[0]]
Expand Down Expand Up @@ -8611,29 +8631,11 @@ def transform_input_spec(self, input_spec):
num_intervals = int(num_intervals.squeeze())
self.num_intervals = torch.as_tensor(num_intervals)

def custom_arange(nint):
result = torch.arange(
start=0.0,
end=1.0,
step=1 / nint,
dtype=self.dtype,
device=action_spec.device,
)
result_ = result
if self.sampling in (
self.SamplingStrategy.HIGH,
self.SamplingStrategy.MEDIAN,
):
result_ = (1 - result).flip(0)
if self.sampling == self.SamplingStrategy.MEDIAN:
result = (result + result_) / 2
else:
result = result_
return result

if isinstance(num_intervals, int):
arange = (
custom_arange(num_intervals).expand((*n_act, num_intervals))
self._custom_arange(num_intervals, action_spec.device).expand(
(*n_act, num_intervals)
)
* interval
)
low = action_spec.low
Expand All @@ -8642,7 +8644,7 @@ def custom_arange(nint):
self.register_buffer("intervals", low + arange)
else:
arange = [
custom_arange(_num_intervals) * interval
self._custom_arange(_num_intervals, action_spec.device) * interval
for _num_intervals, interval in zip(
num_intervals.tolist(), interval.unbind(-2)
)
Expand Down
Loading