[BUG] Same type(x) is tuple anti-pattern as #4372 in gradient/deep explainers and partial_dependence plot #4480

Context

#4372 fixed data is dict in shap/maskers/_tabular.py and shap/explainers/_explainer.py (PRs #4373 / #4424). The same identity-check-against-type anti-pattern exists in four other files that were not covered by those PRs. type(x) is Y compares the exact concrete type and silently rejects subclasses — most notably, NamedTuple instances from PyTorch forward hooks (which are a common input to the gradient/deep explainers).

Locations

14 occurrences across 4 files, verified against current master:

shap/explainers/_gradient.py (4)

• L75 — if type(model) is tuple:
• L520 — if type(interim_inputs) is tuple:
• L667 — if type(interim_inputs) is tuple:
• L668 — if type(interim_inputs) is tuple: ← duplicate of L667; inner branch is either unreachable dead code or masking a typo

shap/explainers/_deep/__init__.py (1)

• L72 — if type(model) is tuple:

shap/explainers/_deep/deep_pytorch.py (4)

• L41 — if type(interim_inputs) is tuple:
• L282 — if i != 0 and type(output) is tuple:
• L288 — if type(input) is tuple:
• L292 — if type(output) is tuple:

shap/plots/_partial_dependence.py (5)

• L71 — if type(ind) is not tuple:
• L216 — xmin0 = xmin[0] if type(xmin) is tuple else xmin
• L217 — xmin1 = xmin[1] if type(xmin) is tuple else xmin
• L218 — xmax0 = xmax[0] if type(xmax) is tuple else xmax
• L219 — xmax1 = xmax[1] if type(xmax) is tuple else xmax

Why it matters

1. User-visible breakage: PyTorch forward hooks can return torch.return_types.* objects, which are tuple subclasses rather than tuple instances — these fail type(x) is tuple. Same for namedtuple/NamedTuple used by third-party models. The gradient and deep explainers silently take the wrong branch.
2. Latent bug at _gradient.py:667-668: Two nested if type(interim_inputs) is tuple: checks. Either the inner block is dead code, or it was meant to check something else (e.g. len(interim_inputs) > 0) and was accidentally duplicated. Worth inspecting before blindly replacing.
3. Consistency with BUG: Type check uses is instead of isinstance in Impute class #4372 direction: The fix there used isinstance(data, dict). Applying the same pattern here keeps the codebase consistent.

Proposed fix

Replace all 14 occurrences with isinstance(x, tuple) / not isinstance(x, tuple). For the duplicated pair at _gradient.py:667-668, investigate git history to determine the original intent and fix accordingly, do not blindly delete the inner block.

I'd like to take this on as a single scoped PR (all four files, one concern: replacing the anti-pattern). Happy to split into one PR per file if that's easier to review.

AI disclosure

Investigation and drafting assisted by Claude. All locations and code citations verified manually against master.