[BugFix] Sample every window in RandomCropTensorDict

[jackiectl2]

Description

RandomCropTensorDict samples the window start with torch.randint(max_idx_0, ...), which draws from [0, max_idx_0). The last valid window is never sampled, and when the sequence length equals sub_seq_len (max_idx_0 == 0) the call fails with RuntimeError: random_ expects 'from' to be less than 'to', although the length check accepts that case. The masked path has the same off-by-one (floor(rand * (L - sub_seq_len)) never reaches L - sub_seq_len).

This samples the start from [0, max_idx_0] (and [0, L - sub_seq_len] with a mask).

import torch
from tensordict import TensorDict
from torchrl.envs.transforms import RandomCropTensorDict

td = TensorDict({"a": torch.arange(4).expand(1000, 4)}, [1000, 4])
print(set(RandomCropTensorDict(2)(td)["a"][:, 0].tolist()))  # main: {0, 1}; expected {0, 1, 2}
RandomCropTensorDict(4)(td)  # main: RuntimeError

Motivation and Context

close #4577

• I have raised an issue to propose this change (required for new features and bug fixes)

Types of changes

• Bug fix (non-breaking change which fixes an issue)

Checklist

• I have read the CONTRIBUTION guide (required)
• I have updated the tests accordingly (required for a bug fix or a new feature).

Tests: added test_crop_full_length and test_crop_samples_every_window (with and without mask) to TestCroSeq; they fail on main and pass with the fix. pytest test/transforms/test_compose_and_env.py: 173 passed, 21 skipped (CPU). pre-commit passes on the changed files.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/4578

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure

As of commit d6810d0 with merge base 264a3c2 ():

NEW FAILURE - The following job has failed:

• Generate documentation / build-docs (3.12, 12.8) / linux-job (gh)
  Could not load credentials from any providers

This comment was automatically generated by Dr. CI and updates every 15 minutes.