Revert back to allowing inducing point method to be optional (faceboo…

…kresearch#486)

Summary:
Pull Request resolved: facebookresearch#486

Inducing point methods can now be initialized within models again with them knowing their own dimensionality. So we allow the inducing point args to be completely optional again. This also lets us reorder the args to be in a more sensible order.

Differential Revision: D67226225

aepsych/models/gp_classification.py

@@ -50,25 +50,26 @@ class GPClassificationModel(AEPsychModelDeviceMixin, ApproximateGP):
     def __init__(
         self,
         dim: int,
-        inducing_point_method: InducingPointAllocator,
-        inducing_size: int = 100,
         mean_module: Optional[gpytorch.means.Mean] = None,
         covar_module: Optional[gpytorch.kernels.Kernel] = None,
         likelihood: Optional[Likelihood] = None,
+        inducing_point_method: Optional[InducingPointAllocator] = None,
+        inducing_size: int = 100,
         max_fit_time: Optional[float] = None,
         optimizer_options: Optional[Dict[str, Any]] = None,
     ) -> None:
         """Initialize the GP Classification model
 
         Args:
             dim (int): The number of dimensions in the parameter space.
-            inducing_point_method (InducingPointAllocator): The method to use for selecting inducing points.
-            inducing_size (int): Number of inducing points. Defaults to 100.
             mean_module (gpytorch.means.Mean, optional): GP mean class. Defaults to a constant with a normal prior.
             covar_module (gpytorch.kernels.Kernel, optional): GP covariance kernel class. Defaults to scaled RBF with a
                 gamma prior.
             likelihood (gpytorch.likelihood.Likelihood, optional): The likelihood function to use. If None defaults to
                 Bernouli likelihood.
+            inducing_point_method (InducingPointAllocator, optional): The method to use for selecting inducing points.
+                If not set, an AutoAllocator is made.
+            inducing_size (int): Number of inducing points. Defaults to 100.
             max_fit_time (float, optional): The maximum amount of time, in seconds, to spend fitting the model. If None,
                 there is no limit to the fitting time.
             optimizer_options (Dict[str, Any], optional): Optimizer options to pass to the SciPy optimizer during
@@ -99,7 +100,9 @@ def __init__(
                 dim=self.dim, stimuli_per_trial=self.stimuli_per_trial
             )
 
-        self.inducing_point_method = inducing_point_method
+        self.inducing_point_method = inducing_point_method or AutoAllocator(
+            dim=self.dim
+        )
         inducing_points = select_inducing_points(
             allocator=self.inducing_point_method,
             inducing_size=self.inducing_size,
@@ -342,31 +345,32 @@ class GPBetaRegressionModel(GPClassificationModel):
     def __init__(
         self,
         dim: int,
-        inducing_point_method: InducingPointAllocator,
-        inducing_size: int = 100,
         mean_module: Optional[gpytorch.means.Mean] = None,
         covar_module: Optional[gpytorch.kernels.Kernel] = None,
         likelihood: Optional[Likelihood] = None,
+        inducing_point_method: Optional[InducingPointAllocator] = None,
+        inducing_size: int = 100,
         max_fit_time: Optional[float] = None,
         optimizer_options: Optional[Dict[str, Any]] = None,
     ) -> None:
         """Initialize the GP Beta Regression model
 
         Args:
             dim (int): The number of dimensions in the parameter space.
-            inducing_point_method (InducingPointAllocator): The method to use to select the inducing points.
-            inducing_size (int, optional): Number of inducing points. Defaults to 100.
             mean_module (gpytorch.means.Mean, optional): GP mean class. Defaults to a constant with a normal prior. Defaults to None.
             covar_module (gpytorch.kernels.Kernel, optional): GP covariance kernel class. Defaults to scaled RBF with a
                 gamma prior.
             likelihood (gpytorch.likelihood.Likelihood, optional): The likelihood function to use. If None defaults to
                 Beta likelihood.
+            inducing_point_method (InducingPointAllocator, optional): The method to use for selecting inducing points.
+                If not set, an AutoAllocator is made.
+            inducing_size (int): Number of inducing points. Defaults to 100.
             max_fit_time (float, optional): The maximum amount of time, in seconds, to spend fitting the model. If None,
                 there is no limit to the fitting time. Defaults to None.
         """
         if likelihood is None:
             likelihood = BetaLikelihood()
-        self.inducing_point_method = inducing_point_method
+
         super().__init__(
             dim=dim,
             mean_module=mean_module,

aepsych/models/gp_regression.py

@@ -18,7 +18,6 @@
 from aepsych.models.utils import select_inducing_points
 from aepsych.utils import get_optimizer_options, promote_0d
 from aepsych.utils_logging import getLogger
-from botorch.models.utils.inducing_point_allocators import InducingPointAllocator
 from gpytorch.likelihoods import GaussianLikelihood, Likelihood
 from gpytorch.models import ExactGP
 

aepsych/models/monotonic_projection_gp.py

@@ -98,13 +98,13 @@ def __init__(
         lb: torch.Tensor,
         ub: torch.Tensor,
         dim: int,
-        inducing_point_method: InducingPointAllocator,
         monotonic_dims: List[int],
         monotonic_grid_size: int = 20,
         min_f_val: Optional[float] = None,
         mean_module: Optional[gpytorch.means.Mean] = None,
         covar_module: Optional[gpytorch.kernels.Kernel] = None,
         likelihood: Optional[Likelihood] = None,
+        inducing_point_method: Optional[InducingPointAllocator] = None,
         inducing_size: int = 100,
         max_fit_time: Optional[float] = None,
         optimizer_options: Optional[Dict[str, Any]] = None,
@@ -115,7 +115,6 @@ def __init__(
             lb (torch.Tensor): Lower bounds of the parameters.
             ub (torch.Tensor): Upper bounds of the parameters.
             dim (int, optional): The number of dimensions in the parameter space.
-            inducing_point_method (InducingPointAllocator): The method for allocating inducing points.
             monotonic_dims (List[int]): A list of the dimensions on which monotonicity should
                 be enforced.
             monotonic_grid_size (int): The size of the grid, s, in 1. above. Defaults to 20.
@@ -125,6 +124,8 @@ def __init__(
                 gamma prior. Defaults to None.
             likelihood (Likelihood, optional): The likelihood function to use. If None defaults to
                 Gaussian likelihood. Defaults to None.
+            inducing_point_method (InducingPointAllocator, optional): The method to use for selecting inducing points.
+                If not set, an AutoAllocator is made.
             inducing_size (int): The number of inducing points to use. Defaults to 100.
             max_fit_time (float, optional): The maximum amount of time, in seconds, to spend fitting the model. If None,
                 there is no limit to the fitting time. Defaults to None.
@@ -133,7 +134,6 @@ def __init__(
         self.monotonic_dims = [int(d) for d in monotonic_dims]
         self.mon_grid_size = monotonic_grid_size
         self.min_f_val = min_f_val
-        self.inducing_point_method = inducing_point_method
         self.lb = lb
         self.ub = ub
 

aepsych/models/semi_p.py

@@ -254,29 +254,30 @@ class SemiParametricGPModel(GPClassificationModel):
     def __init__(
         self,
         dim: int,
-        inducing_point_method: InducingPointAllocator,
-        inducing_size: int = 100,
         stim_dim: int = 0,
         mean_module: Optional[gpytorch.means.Mean] = None,
         covar_module: Optional[gpytorch.kernels.Kernel] = None,
         likelihood: Optional[Any] = None,
         slope_mean: float = 2,
+        inducing_point_method: Optional[InducingPointAllocator] = None,
+        inducing_size: int = 100,
         max_fit_time: Optional[float] = None,
         optimizer_options: Optional[Dict[str, Any]] = None,
     ) -> None:
         """
         Initialize SemiParametricGP.
         Args:
             dim (int, optional): The number of dimensions in the parameter space.
-            inducing_point_method (InducingPointAllocator): The method to use to select the inducing points.
-            inducing_size (int): Number of inducing points. Defaults to 100.
             stim_dim (int): Index of the intensity (monotonic) dimension. Defaults to 0.
             mean_module (gpytorch.means.Mean, optional): GP mean class. Defaults to a constant with a normal prior.
             covar_module (gpytorch.kernels.Kernel, optional): GP covariance kernel class. Defaults to scaled RBF with a
                 gamma prior.
             likelihood (gpytorch.likelihood.Likelihood, optional): The likelihood function to use. If None defaults to
                 linear-Bernouli likelihood with probit link.
             slope_mean (float): The mean of the slope. Defaults to 2.
+            inducing_point_method (InducingPointAllocator, optional): The method to use for selecting inducing points.
+                If not set, an AutoAllocator is made.
+            inducing_size (int): Number of inducing points. Defaults to 100.
             max_fit_time (float, optional): The maximum amount of time, in seconds, to spend fitting the model. If None,
                 there is no limit to the fitting time.
             optimizer_options (Dict[str, Any], optional): Optimizer options to pass to the SciPy optimizer during
@@ -310,7 +311,6 @@ def __init__(
         assert isinstance(
             likelihood, LinearBernoulliLikelihood
         ), "SemiP model only supports linear Bernoulli likelihoods!"
-        self.inducing_point_method = inducing_point_method
 
         super().__init__(
             dim=dim,
@@ -514,37 +514,37 @@ class HadamardSemiPModel(GPClassificationModel):
     def __init__(
         self,
         dim: int,
-        inducing_point_method: InducingPointAllocator,
-        inducing_size: int = 100,
         stim_dim: int = 0,
         slope_mean_module: Optional[gpytorch.means.Mean] = None,
         slope_covar_module: Optional[gpytorch.kernels.Kernel] = None,
         offset_mean_module: Optional[gpytorch.means.Mean] = None,
         offset_covar_module: Optional[gpytorch.kernels.Kernel] = None,
         likelihood: Optional[Likelihood] = None,
         slope_mean: float = 2,
+        inducing_point_method: Optional[InducingPointAllocator] = None,
+        inducing_size: int = 100,
         max_fit_time: Optional[float] = None,
         optimizer_options: Optional[Dict[str, Any]] = None,
     ) -> None:
         """
         Initialize HadamardSemiPModel.
         Args:
             dim (int): The number of dimensions in the parameter space.
-            inducing_point_method (InducingPointAllocator): The method to use to select the inducing points.
-            inducing_size (int): Number of inducing points. Defaults to 100.
             stim_dim (int): Index of the intensity (monotonic) dimension. Defaults to 0.
             slope_mean_module (gpytorch.means.Mean, optional): Mean module to use (default: constant mean) for slope.
             slope_covar_module (gpytorch.kernels.Kernel, optional): Covariance kernel to use (default: scaled RBF) for slope.
             offset_mean_module (gpytorch.means.Mean, optional): Mean module to use (default: constant mean) for offset.
             offset_covar_module (gpytorch.kernels.Kernel, optional): Covariance kernel to use (default: scaled RBF) for offset.
             likelihood (gpytorch.likelihood.Likelihood, optional)): defaults to bernoulli with logistic input and a floor of .5
             slope_mean (float): The mean of the slope. Defaults to 2.
+            inducing_point_method (InducingPointAllocator, optional): The method to use for selecting inducing points.
+                If not set, an AutoAllocator is made.
+            inducing_size (int): Number of inducing points. Defaults to 100.
             max_fit_time (float, optional): The maximum amount of time, in seconds, to spend fitting the model. If None,
                 there is no limit to the fitting time.
             optimizer_options (Dict[str, Any], optional): Optimizer options to pass to the SciPy optimizer during
                 fitting. Assumes we are using L-BFGS-B.
         """
-        self.inducing_point_method = inducing_point_method
         super().__init__(
             dim=dim,
             inducing_size=inducing_size,

tests/test_points_allocators.py

@@ -430,6 +430,12 @@ def test_select_inducing_points_legacy(self):
             )
             self.assertEqual(points.shape, (5, 1))
 
+    def test_model_default_allocator(self):
+        model = GPClassificationModel(dim=2)
+
+        self.assertIsInstance(model.inducing_point_method, AutoAllocator)
+        self.assertTrue(model.inducing_point_method.dim == 2)
+
 
 if __name__ == "__main__":
     unittest.main()