remove input transform checks (#1568)

Summary:
<!--
Thank you for sending the PR! We appreciate you spending the time to make BoTorch better.

Help us understand your motivation by explaining why you decided to make this change.

You can learn more about contributing to BoTorch here: https://github.com/pytorch/botorch/blob/main/CONTRIBUTING.md
-->

## Motivation

As I am currently refactoring our internal codebase, I had a look at sdaulton PR regarding probabilistic reparameterization.
From my understanding one has to use it by representing the categoricals by a one hot encoding for the reparmeterized ACQF and then eventually transforming the input to a numerical represenation via `OneHotToNumeric` especially when one wants to use it togehter with `MixedSingleTaskGP`. Currently MixedSingleTaskGP is very strict on which input transforms are allowed. This PR lifts the restrictions to make it usable with  OneHotToNumeric`.

Note that the transform also has to be instantiated with `transform_on_train = False` and `train_X` has to be transformed before it is passed to the constructor of `MixedSingleTaskGP`, else the indices for the different kernels are mixed up.

### Have you read the [Contributing Guidelines on pull requests](https://github.com/pytorch/botorch/blob/main/CONTRIBUTING.md#pull-requests)?

Yes.

Pull Request resolved: #1568

Test Plan:
Unit tests.

## Related PRs

#1534

Reviewed By: esantorella

Differential Revision: D42230252

Pulled By: Balandat

fbshipit-source-id: b6a0a12d926fbab9890a75438eb60ef849441149

Author: jduerholt

botorch/models/gp_regression_mixed.py

@@ -89,29 +89,17 @@ def __init__(
                 `.posterior` on the model will be on the original scale).
             input_transform: An input transform that is applied in the model's
                 forward pass. Only input transforms are allowed which do not
-                transform the categorical dimensions. This can be achieved
-                by using the `indices` argument when constructing the transform.
+                transform the categorical dimensions. If you want to use it
+                for example in combination with a `OneHotToNumeric` input transform
+                one has to instantiate the transform with `transform_on_train` == False
+                and pass in the already transformed input.
         """
-        if input_transform is not None:
-            if not hasattr(input_transform, "indices"):
-                raise ValueError(
-                    "Only continuous inputs can be transformed. "
-                    "Please use `indices` in the `input_transform`."
-                )
-            # check that no cat dim is in indices
-            elif any(idx in input_transform.indices for idx in cat_dims):
-                raise ValueError(
-                    "Only continuous inputs can be transformed. "
-                    "Categorical index found in `indices` of the `input_transform`."
-                )
         if len(cat_dims) == 0:
             raise ValueError(
                 "Must specify categorical dimensions for MixedSingleTaskGP"
             )
         self._ignore_X_dims_scaling_check = cat_dims
-        input_batch_shape, aug_batch_shape = self.get_batch_dimensions(
-            train_X=train_X, train_Y=train_Y
-        )
+        _, aug_batch_shape = self.get_batch_dimensions(train_X=train_X, train_Y=train_Y)
 
         if cont_kernel_factory is None:
 

test/models/test_gp_regression_mixed.py

@@ -5,7 +5,6 @@
 # LICENSE file in the root directory of this source tree.
 
 import itertools
-import random
 import warnings
 
 import torch
@@ -44,15 +43,6 @@ def test_gp(self):
             )
             cat_dims = list(range(ncat))
             ord_dims = sorted(set(range(d)) - set(cat_dims))
-            with self.assertRaises(ValueError):
-                MixedSingleTaskGP(
-                    train_X,
-                    train_Y,
-                    cat_dims=cat_dims,
-                    input_transform=Normalize(
-                        d=d, bounds=bounds.to(**tkwargs), transform_on_train=True
-                    ),
-                )
             # test correct indices
             if (ncat < 3) and (ncat > 0):
                 MixedSingleTaskGP(
@@ -66,30 +56,6 @@ def test_gp(self):
                         indices=ord_dims,
                     ),
                 )
-                with self.assertRaises(ValueError):
-                    MixedSingleTaskGP(
-                        train_X,
-                        train_Y,
-                        cat_dims=cat_dims,
-                        input_transform=Normalize(
-                            d=d,
-                            bounds=bounds.to(**tkwargs),
-                            transform_on_train=True,
-                            indices=cat_dims,
-                        ),
-                    )
-                with self.assertRaises(ValueError):
-                    MixedSingleTaskGP(
-                        train_X,
-                        train_Y,
-                        cat_dims=cat_dims,
-                        input_transform=Normalize(
-                            d=d,
-                            bounds=bounds.to(**tkwargs),
-                            transform_on_train=True,
-                            indices=ord_dims + [random.choice(cat_dims)],
-                        ),
-                    )
 
             if len(cat_dims) == 0:
                 with self.assertRaises(ValueError):