introduce trial_indices argument to SupervisedDataset (pytorch#2595)

Summary:
X-link: facebook/Ax#2960


Adds optional `trial_indices` to SupervisedDataset, whose dimensionality should correspond 1:1 with the first few dimensions of X and Y tensors, as validated in `_validate` ([pointer](https://www.internalfb.com/diff/D64764019?permalink=1739375523489084)).

Reviewed By: Balandat

Differential Revision: D64764019

botorch/utils/datasets.py

@@ -14,6 +14,7 @@
 import torch
 from botorch.exceptions.errors import InputDataError, UnsupportedError
 from botorch.utils.containers import BotorchContainer, SliceContainer
+from pyre_extensions import none_throws
 from torch import long, ones, Tensor
 
 
@@ -54,6 +55,7 @@ def __init__(
         outcome_names: list[str],
         Yvar: BotorchContainer | Tensor | None = None,
         validate_init: bool = True,
+        trial_indices: Tensor | None = None,
     ) -> None:
         r"""Constructs a `SupervisedDataset`.
 
@@ -65,12 +67,16 @@ def __init__(
             Yvar: An optional `Tensor` or `BotorchContainer` representing
                 the observation noise.
             validate_init: If `True`, validates the input shapes.
+            trial_indices: A `Tensor` representing the trial indices of X and Y. This is
+                used to support learning-curve-based modeling. If provided, it must
+                have compatible shape with X and Y.
         """
         self._X = X
         self._Y = Y
         self._Yvar = Yvar
         self.feature_names = feature_names
         self.outcome_names = outcome_names
+        self.trial_indices = trial_indices
         if validate_init:
             self._validate()
 
@@ -96,6 +102,7 @@ def _validate(
         self,
         validate_feature_names: bool = True,
         validate_outcome_names: bool = True,
+        validate_trial_indices: bool = True,
     ) -> None:
         r"""Checks that the shapes of the inputs are compatible with each other.
 
@@ -108,6 +115,8 @@ def _validate(
                 `outcomes_names` matches the # of columns of `self.Y`. If a
                 particular dataset, e.g., `RankingDataset`, is known to violate
                 this assumption, this can be set to `False`.
+            validate_trial_indices: By default, we validate that the shape of
+                `trial_indices` matches the shape of X and Y.
         """
         shape_X = self.X.shape
         if isinstance(self._X, BotorchContainer):
@@ -133,8 +142,19 @@ def _validate(
                 "`Y` must have the same number of columns as the number of "
                 "outcomes in `outcome_names`."
             )
+        if validate_trial_indices and self.trial_indices is not None:
+            if self.trial_indices.shape != shape_X:
+                raise ValueError(
+                    f"{shape_X=} must have the same shape as {none_throws(self.trial_indices).shape=}."
+                )
 
     def __eq__(self, other: Any) -> bool:
+        if self.trial_indices is None and other.trial_indices is None:
+            trial_indices_equal = True
+        elif self.trial_indices is None or other.trial_indices is None:
+            trial_indices_equal = False
+        else:
+            trial_indices_equal = torch.equal(self.trial_indices, other.trial_indices)
         return (
             type(other) is type(self)
             and torch.equal(self.X, other.X)
@@ -146,6 +166,7 @@ def __eq__(self, other: Any) -> bool:
             )
             and self.feature_names == other.feature_names
             and self.outcome_names == other.outcome_names
+            and trial_indices_equal
         )
 
 
@@ -241,7 +262,11 @@ def __init__(
         )
 
     def _validate(self) -> None:
-        super()._validate(validate_feature_names=False, validate_outcome_names=False)
+        super()._validate(
+            validate_feature_names=False,
+            validate_outcome_names=False,
+            validate_trial_indices=False,
+        )
         if len(self.feature_names) != self._X.values.shape[-1]:
             raise ValueError(
                 "The `values` field of `X` must have the same number of columns as "
@@ -316,6 +341,7 @@ def __init__(
         self.has_heterogeneous_features = any(
             datasets[0].feature_names != ds.feature_names for ds in datasets[1:]
         )
+        self.trial_indices = None
 
     @classmethod
     def from_joint_dataset(
@@ -538,6 +564,7 @@ def __init__(
             c: [self.feature_names.index(i) for i in parameter_decomposition[c]]
             for c in self.context_buckets
         }
+        self.trial_indices = None
 
     @property
     def X(self) -> Tensor:

test/utils/test_datasets.py

@@ -43,27 +43,35 @@ def make_dataset(
 class TestDatasets(BotorchTestCase):
     def test_supervised(self):
         # Generate some data
-        X = rand(3, 2)
-        Y = rand(3, 1)
+        n_rows = 3
+        X = rand(n_rows, 2)
+        Y = rand(n_rows, 1)
         feature_names = ["x1", "x2"]
         outcome_names = ["y"]
+        trial_indices = tensor(range(n_rows))
 
         # Test `__init__`
         dataset = SupervisedDataset(
-            X=X, Y=Y, feature_names=feature_names, outcome_names=outcome_names
+            X=X,
+            Y=Y,
+            feature_names=feature_names,
+            outcome_names=outcome_names,
+            trial_indices=trial_indices,
         )
         self.assertIsInstance(dataset.X, Tensor)
         self.assertIsInstance(dataset._X, Tensor)
         self.assertIsInstance(dataset.Y, Tensor)
         self.assertIsInstance(dataset._Y, Tensor)
         self.assertEqual(dataset.feature_names, feature_names)
         self.assertEqual(dataset.outcome_names, outcome_names)
+        self.assertTrue(torch.equal(dataset.trial_indices, trial_indices))
 
         dataset2 = SupervisedDataset(
             X=DenseContainer(X, X.shape[-1:]),
             Y=DenseContainer(Y, Y.shape[-1:]),
             feature_names=feature_names,
             outcome_names=outcome_names,
+            trial_indices=trial_indices,
         )
         self.assertIsInstance(dataset2.X, Tensor)
         self.assertIsInstance(dataset2._X, DenseContainer)
@@ -101,6 +109,14 @@ def test_supervised(self):
                 feature_names=feature_names,
                 outcome_names=[],
             )
+        with self.assertRaisesRegex(ValueError, "trial_indices"):
+            SupervisedDataset(
+                X=rand(2, 2),
+                Y=rand(2, 1),
+                feature_names=feature_names,
+                outcome_names=outcome_names,
+                trial_indices=tensor(range(n_rows + 1)),
+            )
 
         # Test with Yvar.
         dataset = SupervisedDataset(