[BugFix] Deprecate tensordict.set check skips in transforms (#951)

Author: vmoens

torchrl/envs/common.py

@@ -830,7 +830,6 @@ def fake_tensordict(self) -> TensorDictBase:
             },
             batch_size=self.batch_size,
             device=self.device,
-            _run_checks=True,  # this method should not be run very often. This facilitates debugging
         )
         return fake_td
 

torchrl/envs/transforms/transforms.py

@@ -2638,18 +2638,27 @@ def _update(self, key, value, N) -> torch.Tensor:
         value_sum = _sum_left(value, _sum)
         _sum *= self.decay
         _sum += value_sum
-        self._td.set_(key + "_sum", _sum, no_check=True)
+        self._td.set_(
+            key + "_sum",
+            _sum,
+        )
 
         _ssq = self._td.get(key + "_ssq")
         value_ssq = _sum_left(value.pow(2), _ssq)
         _ssq *= self.decay
         _ssq += value_ssq
-        self._td.set_(key + "_ssq", _ssq, no_check=True)
+        self._td.set_(
+            key + "_ssq",
+            _ssq,
+        )
 
         _count = self._td.get(key + "_count")
         _count *= self.decay
         _count += N
-        self._td.set_(key + "_count", _count, no_check=True)
+        self._td.set_(
+            key + "_count",
+            _count,
+        )
 
         mean = _sum / _count
         std = (_ssq / _count - mean.pow(2)).clamp_min(self.eps).sqrt()

torchrl/envs/utils.py

@@ -27,7 +27,6 @@ def step_mdp(
     exclude_reward: bool = True,
     exclude_done: bool = True,
     exclude_action: bool = True,
-    _run_check: bool = True,
 ) -> TensorDictBase:
     """Creates a new tensordict that reflects a step in time of the input tensordict.