test: pin TFTInstanceSplitter observed_value_field=None branch (#3296 item 3)

[jbbqqf]

Issue #, if available: refs #3296 (item 3)

Description of changes:

Summary

Add a regression test pinning the observed_value_field=None branch of TFTInstanceSplitter, introduced by #3259 but not covered by the test suite.

Context

#3296 enumerates regression-test gaps left by recent merged PRs. Item 3:

#3259 — observed_value_field: Optional[str]: no test passing None through TFTInstanceSplitter on a toy dataset. The happy path with the default field name is covered; the new None branch isn't.

grep -rn TFTInstanceSplitter test/ confirms there were zero direct tests for this class. Without a test, a future refactor could regress the None branch silently — the flatmap_transform method only branches on if self.observed_value_field is not None: (src/gluonts/transform/split.py:533-534), and the integration tests in test/torch/model/test_estimators.py only exercise the default field-name path.

Changes

• test/transform/test_transform.py: add test_TFTInstanceSplitter_observed_value_field_optional, a pytest.parametrize over [None, FieldName.OBSERVED_VALUES] that:
  • Runs the splitter end-to-end on a small data dict using TestSplitSampler.
  • Asserts past_target always has the configured past_length.
  • Asserts the None branch emits neither observed_values nor past_observed_values, and does not require the entry to carry an observed-values array.
  • Asserts the default-like branch emits past_observed_values with the right length.

A short docstring on the test cites #3259 and #3296 so a reviewer reading the test cold sees the regression it guards.

Reproduce BEFORE/AFTER yourself (copy-paste)

git clone https://github.com/awslabs/gluonts.git /tmp/repro && cd /tmp/repro
python3.11 -m venv .venv && source .venv/bin/activate
pip install -e . pytest

# --- BEFORE (origin/dev) ---
git checkout origin/dev
pytest test/transform/test_transform.py -q -k TFTInstanceSplitter
# Expected: 0 tests collected — there is no TFTInstanceSplitter test on dev.

# --- AFTER (this PR) ---
git fetch https://github.com/jbbqqf/gluonts.git feat/3296a-tft-splitter-test
git checkout FETCH_HEAD
pytest test/transform/test_transform.py -q -k TFTInstanceSplitter
# Expected: 2 passed — both branches pinned.

What I ran locally

• pytest test/transform/test_transform.py -q → 252 passed, 4 skipped on the branch (was 250 + 4 on dev; +2 from this PR).
• pytest test/transform/test_transform.py -q -k TFTInstanceSplitter → 2 passed.

Edge cases tested

#	Scenario	Input	Expected	Verified by
1	observed_value_field=None	data dict without an observed-values array	splitter returns one entry; no observed_values / past_observed_values keys	test_TFTInstanceSplitter_observed_value_field_optional[None]
2	observed_value_field=FieldName.OBSERVED_VALUES	data dict with observed_values mask	splitter returns one entry with past_observed_values of length past_length	test_TFTInstanceSplitter_observed_value_field_optional[observed_values]

Risk / blast radius

Test-only addition. Zero runtime impact.

Release note

NONE

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

Please tag this pr with at least one of these labels to make our release process faster: BREAKING, new feature, bug fix, other change, dev setup

---

PR drafted with assistance from Claude Code. The reproducer block above was used during development and is the same one a reviewer can paste verbatim.