Short-cut t_batch_mode_transform decorator on non-tensor inputs (#991)

Summary:
This essentially makes `t_batch_mode_tarnsform` a nullop in case the argument `X` passed to the acquisition function is not a `torch.Tensor` object. This allows using acquisition functions that use models with non non-standard input types such as strings, which is the case in some applications.

Currently this just touches the decorator; in the future we should consider changing the types and signatures of the acquisition functions and models throughout to natively support this more generally.

cc wjmaddox

Pull Request resolved: #991

Reviewed By: dme65

Differential Revision: D32903859

Pulled By: Balandat

fbshipit-source-id: c3abc8b40db307358807fe60014e2c0e1fe49c58

Author: Balandat

botorch/utils/transforms.py

@@ -12,11 +12,13 @@
 
 import warnings
 from functools import wraps
-from typing import Any, Callable, List, Optional
+from typing import Any, Callable, List, Optional, TypeVar
 
 import torch
 from torch import Tensor
 
+ACQF = TypeVar("AcquisitionFunction")
+
 
 def squeeze_last_dim(Y: Tensor) -> Tensor:
     r"""Squeeze the last dimension of a Tensor.
@@ -172,19 +174,19 @@ def _verify_output_shape(acqf: Any, X: Tensor, output: Tensor) -> bool:
 def t_batch_mode_transform(
     expected_q: Optional[int] = None,
     assert_output_shape: bool = True,
-) -> Callable[[Callable[[Any, Tensor], Any]], Callable[[Any, Tensor], Any]]:
-    r"""Factory for decorators taking a t-batched `X` tensor.
+) -> Callable[[Callable[[ACQF, Any], Any]], Callable[[ACQF, Any], Any]]:
+    r"""Factory for decorators enabling consistent t-batch behavior.
 
     This method creates decorators for instance methods to transform an input tensor
     `X` to t-batch mode (i.e. with at least 3 dimensions). This assumes the tensor
     has a q-batch dimension. The decorator also checks the q-batch size if `expected_q`
     is provided, and the output shape if `assert_output_shape` is `True`.
 
     Args:
-        expected_q: The expected q-batch size of X. If specified, this will raise an
-            AssertionError if X's q-batch size does not equal expected_q.
+        expected_q: The expected q-batch size of `X`. If specified, this will raise an
+            AssertionError if `X`'s q-batch size does not equal expected_q.
         assert_output_shape: If `True`, this will raise an AssertionError if the
-            output shape does not match either the t-batch shape of X,
+            output shape does not match either the t-batch shape of `X`,
             or the `acqf.model.batch_shape` for acquisition functions using
             batched models.
 
@@ -202,9 +204,16 @@ def t_batch_mode_transform(
         >>>         ...
     """
 
-    def decorator(method: Callable[[Any, Tensor], Any]) -> Callable[[Any, Tensor], Any]:
+    def decorator(
+        method: Callable[[ACQF, Any], Any],
+    ) -> Callable[[ACQF, Any], Any]:
         @wraps(method)
-        def decorated(acqf: Any, X: Tensor, *args: Any, **kwargs: Any) -> Any:
+        def decorated(acqf: ACQF, X: Any, *args: Any, **kwargs: Any) -> Any:
+
+            # Allow using acquisition functions for other inputs (e.g. lists of strings)
+            if not isinstance(X, Tensor):
+                return method(acqf, X, *args, **kwargs)
+
             if X.dim() < 2:
                 raise ValueError(
                     f"{type(acqf).__name__} requires X to have at least 2 dimensions,"

test/utils/test_transforms.py

@@ -191,6 +191,11 @@ def test_t_batch_mode_transform(self):
         Xout = c.broadcast_batch_shape_method(X)
         self.assertEqual(Xout.shape, c.model.batch_shape)
 
+        # test with non-tensor argument
+        X = ((3, 4), {"foo": True})
+        Xout = c.q_method(X)
+        self.assertEqual(X, Xout)
+
 
 class TestConcatenatePendingPoints(BotorchTestCase):
     def test_concatenate_pending_points(self):