init

.github/scripts/m1_script.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
 
-export TORCHRL_BUILD_VERSION=0.4.0
+export TORCHRL_BUILD_VERSION=0.5.0
 
 ${CONDA_RUN} pip install git+https://github.com/pytorch/tensordict.git -U

.github/workflows/wheels.yml

@@ -32,7 +32,7 @@ jobs:
         run: |
           export PATH="/opt/python/${{ matrix.python_version[1] }}/bin:$PATH"
           python3 -mpip install wheel
-          TORCHRL_BUILD_VERSION=0.4.0 python3 setup.py bdist_wheel
+          TORCHRL_BUILD_VERSION=0.5.0 python3 setup.py bdist_wheel
           # NB: wheels have the linux_x86_64 tag so we rename to manylinux1
           # find . -name 'dist/*whl' -exec bash -c ' mv $0 ${0/linux/manylinux1}' {} \;
       # pytorch/pytorch binaries are also manylinux_2_17 compliant but they
@@ -72,7 +72,7 @@ jobs:
         shell: bash
         run: |
             python3 -mpip install wheel
-            TORCHRL_BUILD_VERSION=0.4.0 python3 setup.py bdist_wheel
+            TORCHRL_BUILD_VERSION=0.5.0 python3 setup.py bdist_wheel
       - name: Upload wheel for the test-wheel job
         uses: actions/upload-artifact@v2
         with:

setup.py

@@ -172,7 +172,7 @@ def _main(argv):
     if is_nightly:
         tensordict_dep = "tensordict-nightly"
     else:
-        tensordict_dep = "tensordict>=0.4.0"
+        tensordict_dep = "tensordict>=0.5.0"
 
     if is_nightly:
         version = get_nightly_version()

torchrl/collectors/collectors.py

@@ -2065,18 +2065,18 @@ def _queue_len(self) -> int:
     def iterator(self) -> Iterator[TensorDictBase]:
         cat_results = self.cat_results
         if cat_results is None:
-            cat_results = 0
+            cat_results = "stack"
             warnings.warn(
                 f"`cat_results` was not specified in the constructor of {type(self).__name__}. "
                 f"For MultiSyncDataCollector, `cat_results` indicates how the data should "
-                f"be packed: the preferred option is `cat_results='stack'` which provides "
-                f"the best interoperability across torchrl components. "
+                f"be packed: the preferred option and current default is `cat_results='stack'` "
+                f"which provides the best interoperability across torchrl components. "
                 f"Other accepted values are `cat_results=0` (previous behaviour) and "
                 f"`cat_results=-1` (cat along time dimension). Among these two, the latter "
                 f"should be preferred for consistency across environment configurations. "
-                f"Currently, the default value is `0` (using torch.cat along first dimension)."
-                f"From v0.5 onward, this will default to `'stack'`. "
-                f"To suppress this warning, set stack_results to the desired value.",
+                f"Currently, the default value is `'stack'`."
+                f"From v0.6 onward, this warning will be removed. "
+                f"To suppress this warning, set `cat_results` to the desired value.",
                 category=DeprecationWarning,
             )
 

torchrl/envs/common.py

@@ -15,7 +15,6 @@
 import torch
 import torch.nn as nn
 from tensordict import LazyStackedTensorDict, TensorDictBase, unravel_key
-from tensordict.base import NO_DEFAULT
 from tensordict.utils import NestedKey
 from torchrl._utils import (
     _ends_with,
@@ -3020,21 +3019,11 @@ class _EnvWrapper(EnvBase):
     def __init__(
         self,
         *args,
-        device: DEVICE_TYPING = NO_DEFAULT,
+        device: DEVICE_TYPING = None,
         batch_size: Optional[torch.Size] = None,
         allow_done_after_reset: bool = False,
         **kwargs,
     ):
-        if device is NO_DEFAULT:
-            warnings.warn(
-                "Your wrapper was not given a device. Currently, this "
-                "value will default to 'cpu'. From v0.5 it will "
-                "default to `None`. With a device of None, no device casting "
-                "is performed and the resulting tensordicts are deviceless. "
-                "Please set your device accordingly.",
-                category=DeprecationWarning,
-            )
-            device = torch.device("cpu")
         super().__init__(
             device=device,
             batch_size=batch_size,

torchrl/envs/transforms/transforms.py

@@ -3411,14 +3411,7 @@ def __init__(
         out_keys_inv: Sequence[NestedKey] | None = None,
     ):
         if in_keys is not None and in_keys_inv is None:
-            warnings.warn(
-                "in_keys have been provided but not in_keys_inv. From v0.5, "
-                "this will result in in_keys_inv being an empty list whereas "
-                "now the input keys are retrieved automatically. "
-                "To silence this warning, pass the (possibly empty) "
-                "list of in_keys_inv.",
-                category=DeprecationWarning,
-            )
+            in_keys_inv = []
 
         self.dtype_in = dtype_in
         self.dtype_out = dtype_out

torchrl/modules/distributions/continuous.py

@@ -481,9 +481,10 @@ def root_dist(self):
     @property
     def mode(self):
         warnings.warn(
-            "This computation of the mode is based on the first-order Taylor expansion "
-            "of the transform around the normal mean value, which can be inaccurate. "
+            "This computation of the mode is based on an inaccurate estimation of the mode "
+            "given the base_dist mode. "
             "To use a more stable implementation of the mode, use dist.get_mode() method instead. "
+            "To silence this warning, consider using the DETERMINISTIC exploration_type."
             "This implementation will be removed in v0.6.",
             category=DeprecationWarning,
         )

torchrl/modules/tensordict_module/actors.py

@@ -4,7 +4,6 @@
 # LICENSE file in the root directory of this source tree.
 from __future__ import annotations
 
-import warnings
 from typing import Dict, List, Optional, Sequence, Tuple, Union
 
 import torch
@@ -922,10 +921,9 @@ def __init__(
         out_keys: Optional[Sequence[NestedKey]] = None,
     ):
         if isinstance(action_space, TensorSpec):
-            warnings.warn(
-                "Using specs in action_space will be deprecated in v0.4.0,"
-                " please use the 'spec' argument if you want to provide an action spec",
-                category=DeprecationWarning,
+            raise RuntimeError(
+                "Using specs in action_space is deprecated. "
+                "Please use the 'spec' argument if you want to provide an action spec"
             )
         action_space, _ = _process_action_space_spec(action_space, None)
 
@@ -1136,10 +1134,9 @@ def __init__(
         action_mask_key: Optional[NestedKey] = None,
     ):
         if isinstance(action_space, TensorSpec):
-            warnings.warn(
-                "Using specs in action_space will be deprecated v0.4.0,"
-                " please use the 'spec' argument if you want to provide an action spec",
-                category=DeprecationWarning,
+            raise RuntimeError(
+                "Using specs in action_space is deprecated."
+                "Please use the 'spec' argument if you want to provide an action spec"
             )
         action_space, spec = _process_action_space_spec(action_space, spec)
 

version.txt

@@ -1 +1 @@
-0.4.0
+0.5.0