RobustRelevancePursuitSingleTaskGP with specialized `fit_gpytorch_m…

…ll` (pytorch#2690)

Summary:
Pull Request resolved: pytorch#2690

This commit introduces an abstract `RobustRelevancePursuitModel` and `RobustRelevancePursuitSingleTaskGP`, a specific implementation of the abstract class. The main purpose of the new class is to provide an identical interface to a canonical `SingleTaskGP`, but automatically extend the likelihood with the `SparseOutlierGaussianLikelihood`, and toggle the Relevance Pursuit algorithm automatically through the marginal likelihood optimization via `fit_gpytorch_mll` by dispatching on the model type. This makes the model and algorithm easy to use.

Reviewed By: esantorella

Differential Revision: D68353582

botorch/models/gp_regression.py

@@ -186,6 +186,7 @@ def __init__(
                     noise=train_Yvar, batch_shape=self._aug_batch_shape
                 )
         else:
+            # This is used to check if the `model_list_to_batched` can be used
             self._is_custom_likelihood = True
         ExactGP.__init__(
             self, train_inputs=train_X, train_targets=train_Y, likelihood=likelihood

botorch/models/relevance_pursuit.py

@@ -20,7 +20,7 @@
 import math
 
 from abc import ABC, abstractmethod
-from collections.abc import Callable
+from collections.abc import Callable, Sequence
 from copy import copy, deepcopy
 from functools import partial
 from typing import Any, cast, Optional
@@ -35,12 +35,13 @@
 
 MLL_ITER = 10_000  # let's take convergence seriously
 MLL_TOL = 1e-8
-RESET_PARAMETERS = False
+RESET_PARAMETERS = True
+RESET_DENSE_PARAMETERS = False
 
 
 class RelevancePursuitMixin(ABC):
     """Mixin class to convert between the sparse and dense representations of the
-    relevance pursuit models' sparse parameters, as well as to compute the generalized
+    relevance pursuit modules' sparse parameters, as well as to compute the generalized
     support acquisition and support deletion criteria.
     """
 
@@ -251,19 +252,21 @@ def support_expansion(
         n: int = 1,
         modifier: Callable[[Tensor], Tensor] | None = None,
     ) -> bool:
-        """Computes the indices of the features that maximize the gradient of the sparse
+        """Computes the indices of the elements that maximize the gradient of the sparse
         parameter and that are not already in the support, and subsequently expands the
-        support to include the features if their gradient is positive.
+        support to include the elements if their gradient is positive.
 
         Args:
             mll: The marginal likelihood, containing the model to optimize.
                 NOTE: Virtually all of the rest of the code is not specific to the
                 marginal likelihood optimization, so we could generalize this to work
                 with any objective.
-            n: The number of features to select.
-            modifier: A function that modifies the gradient of the inactive parameters
+            n: The maximum number of elements to select. NOTE: The actual number of
+                elements that are added could be fewer if there are fewer than `n`
+                elements with a positive gradient.
+            modifier: A function that modifies the gradient of the inactive elements
                 before computing the support expansion criterion. This can be used
-                to select the maximum gradient magnitude for real-valued parameters
+                to select the maximum gradient magnitude for real-valued elements
                 whose gradients are not non-negative, using modifier = torch.abs.
 
         Returns:
@@ -354,15 +357,15 @@ def support_contraction(
         n: int = 1,
         modifier: Callable[[Tensor], Tensor] | None = None,
     ) -> bool:
-        """Computes the indices of the features that have the smallest coefficients,
-        and subsequently contracts the exlude the features.
+        """Computes the indices of the elements with the smallest magnitude,
+        and subsequently contracts the support by exluding the elements.
 
         Args:
             mll: The marginal likelihood, containing the model to optimize.
                 NOTE: Virtually all of the rest of the code is not specific to the
                 marginal likelihood optimization, so we could generalize this to work
                 with any objective.
-            n: The number of features to select for removal.
+            n: The number of elements to select for removal.
             modifier: A function that modifies the parameter values before computing
                 the support contraction criterion.
 
@@ -395,7 +398,11 @@ def optimize_mll(
         mll: ExactMarginalLogLikelihood,
         model_trace: list[Model] | None = None,
         reset_parameters: bool = RESET_PARAMETERS,
-        reset_dense_parameters: bool = RESET_PARAMETERS,
+        reset_dense_parameters: bool = RESET_DENSE_PARAMETERS,
+        # fit_gpytorch_mll kwargs
+        closure: Callable[[], tuple[Tensor, Sequence[Tensor | None]]] | None = None,
+        optimizer: Callable | None = None,
+        closure_kwargs: dict[str, Any] | None = None,
         optimizer_kwargs: dict[str, Any] | None = None,
     ):
         """Optimizes the marginal likelihood.
@@ -410,6 +417,10 @@ def optimize_mll(
             reset_dense_parameters: If True, re-initializes the dense parameters, e.g.
                 other GP hyper-parameters that are *not* part of the Relevance Pursuit
                 module, to the initial values provided by their associated constraints.
+            closure: A closure to use to compute the loss and the gradients, see
+                docstring of `fit_gpytorch_mll` for details.
+            optimizer: The numerical optimizer, see docstring of `fit_gpytorch_mll`.
+            closure_kwargs: Additional arguments to pass to the `closure` function.
             optimizer_kwargs: A dictionary of keyword arguments for the optimizer.
 
         Returns:
@@ -419,7 +430,6 @@ def optimize_mll(
             # this might be beneficial because the parameters can
             # end up at a constraint boundary, which can anecdotally make
             # it more difficult to move the newly added parameters.
-            # should we only do this after expansion?
             with torch.no_grad():
                 self.sparse_parameter.zero_()
 
@@ -430,7 +440,13 @@ def optimize_mll(
         # NOTE: this function should never force the dense representation, because some
         # models might never need it, and it would be inefficient.
         self.to_sparse()
-        fit_gpytorch_mll(mll, optimizer_kwargs=optimizer_kwargs)
+        mll = fit_gpytorch_mll(
+            mll,
+            optimizer_kwargs=optimizer_kwargs,
+            closure=closure,
+            optimizer=optimizer,
+            closure_kwargs=closure_kwargs,
+        )
         if model_trace is not None:
             # need to record the full model here, rather than just the sparse parameter
             # since other hyper-parameters are co-adapted to the sparse parameter.
@@ -443,11 +459,15 @@ def forward_relevance_pursuit(
     sparse_module: RelevancePursuitMixin,
     mll: ExactMarginalLogLikelihood,
     sparsity_levels: list[int] | None = None,
-    optimizer_kwargs: dict[str, Any] | None = None,
     reset_parameters: bool = RESET_PARAMETERS,
-    reset_dense_parameters: bool = RESET_PARAMETERS,
+    reset_dense_parameters: bool = RESET_DENSE_PARAMETERS,
     record_model_trace: bool = True,
     initial_support: list[int] | None = None,
+    # fit_gpytorch_mll kwargs
+    closure: Callable[[], tuple[Tensor, Sequence[Tensor | None]]] | None = None,
+    optimizer: Callable | None = None,
+    closure_kwargs: dict[str, Any] | None = None,
+    optimizer_kwargs: dict[str, Any] | None = None,
 ) -> tuple[RelevancePursuitMixin, Optional[list[Model]]]:
     """Forward Relevance Pursuit.
 
@@ -478,9 +498,6 @@ def forward_relevance_pursuit(
         sparse_module: The relevance pursuit module.
         mll: The marginal likelihood, containing the model to optimize.
         sparsity_levels: The sparsity levels to expand the support to.
-        optimizer_kwargs: A dictionary of keyword arguments to pass to the optimizer.
-            By default, initializes the "options" sub-dictionary with `maxiter` and
-            `ftol`, `gtol` values, unless specified.
         reset_parameters: If true, initializes the sparse parameter to the all zeros
             after each iteration.
         reset_dense_parameters: If true, re-initializes the dense parameters, e.g.
@@ -489,6 +506,13 @@ def forward_relevance_pursuit(
         record_model_trace: If true, records the model state after every iteration.
         initial_support: The support with which to initialize the sparse module. By
             default, the support is initialized to the empty set.
+        closure: A closure to use to compute the loss and the gradients, see docstring
+            of `fit_gpytorch_mll` for details.
+        optimizer: The numerical optimizer, see docstring of `fit_gpytorch_mll`.
+        closure_kwargs: Additional arguments to pass to the `closure` function.
+        optimizer_kwargs: A dictionary of keyword arguments to pass to the optimizer.
+            By default, initializes the "options" sub-dictionary with `maxiter` and
+            `ftol`, `gtol` values, unless specified.
 
     Returns:
         The relevance pursuit module after forward relevance pursuit optimization, and
@@ -510,14 +534,17 @@ def forward_relevance_pursuit(
 
     model_trace = [] if record_model_trace else None
 
-    def optimize_mll(mll):
-        return sparse_module.optimize_mll(
-            mll=mll,
-            model_trace=model_trace,
-            reset_parameters=reset_parameters,
-            reset_dense_parameters=reset_dense_parameters,
-            optimizer_kwargs=optimizer_kwargs,
-        )
+    optimize_mll = partial(
+        sparse_module.optimize_mll,
+        model_trace=model_trace,
+        reset_parameters=reset_parameters,
+        reset_dense_parameters=reset_dense_parameters,
+        # These are the args of the canonical mll fit routine
+        closure=closure,
+        optimizer=optimizer,
+        closure_kwargs=closure_kwargs,
+        optimizer_kwargs=optimizer_kwargs,
+    )
 
     # if sparsity levels contains the initial support, remove it
     if sparsity_levels[0] == len(sparse_module.support):
@@ -548,11 +575,15 @@ def backward_relevance_pursuit(
     sparse_module: RelevancePursuitMixin,
     mll: ExactMarginalLogLikelihood,
     sparsity_levels: list[int] | None = None,
-    optimizer_kwargs: dict[str, Any] | None = None,
     reset_parameters: bool = RESET_PARAMETERS,
-    reset_dense_parameters: bool = RESET_PARAMETERS,
+    reset_dense_parameters: bool = RESET_DENSE_PARAMETERS,
     record_model_trace: bool = True,
     initial_support: list[int] | None = None,
+    # fit_gpytorch_mll kwargs
+    closure: Callable[[], tuple[Tensor, Sequence[Tensor | None]]] | None = None,
+    optimizer: Callable | None = None,
+    closure_kwargs: dict[str, Any] | None = None,
+    optimizer_kwargs: dict[str, Any] | None = None,
 ) -> tuple[RelevancePursuitMixin, Optional[list[Model]]]:
     """Backward Relevance Pursuit.
 
@@ -583,9 +614,6 @@ def backward_relevance_pursuit(
         sparse_module: The relevance pursuit module.
         mll: The marginal likelihood, containing the model to optimize.
         sparsity_levels: The sparsity levels to expand the support to.
-        optimizer_kwargs: A dictionary of keyword arguments to pass to the optimizer.
-            By default, initializes the "options" sub-dictionary with `maxiter` and
-            `ftol`, `gtol` values, unless specified.
         reset_parameters: If true, initializes the sparse parameter to the all zeros
             after each iteration.
         reset_dense_parameters: If true, re-initializes the dense parameters, e.g.
@@ -594,6 +622,13 @@ def backward_relevance_pursuit(
         record_model_trace: If true, records the model state after every iteration.
         initial_support: The support with which to initialize the sparse module. By
             default, the support is initialized to the full set.
+        closure: A closure to use to compute the loss and the gradients, see docstring
+            of `fit_gpytorch_mll` for details.
+        optimizer: The numerical optimizer, see docstring of `fit_gpytorch_mll`.
+        closure_kwargs: Additional arguments to pass to the `closure` function.
+        optimizer_kwargs: A dictionary of keyword arguments to pass to the optimizer.
+            By default, initializes the "options" sub-dictionary with `maxiter` and
+            `ftol`, `gtol` values, unless specified.
 
     Returns:
         The relevance pursuit module after forward relevance pursuit optimization, and
@@ -623,6 +658,10 @@ def optimize_mll(mll):
             model_trace=model_trace,
             reset_parameters=reset_parameters,
             reset_dense_parameters=reset_dense_parameters,
+            # These are the args of the canonical mll fit routine
+            closure=closure,
+            optimizer=optimizer,
+            closure_kwargs=closure_kwargs,
             optimizer_kwargs=optimizer_kwargs,
         )