move objective outside of acquistion folder

aepsych/acquisition/lse.py

@@ -8,7 +8,7 @@
 from typing import Optional, Union
 
 import torch
-from aepsych.acquisition.objective import ProbitObjective
+from aepsych.objective import ProbitObjective
 from botorch.acquisition.input_constructors import acqf_input_constructor
 from botorch.acquisition.monte_carlo import (
     MCAcquisitionFunction,

aepsych/acquisition/mc_posterior_variance.py

@@ -9,7 +9,7 @@
 
 import torch
 from aepsych.acquisition.monotonic_rejection import MonotonicMCAcquisition
-from aepsych.acquisition.objective import ProbitObjective
+from aepsych.objective import ProbitObjective
 from botorch.acquisition.input_constructors import acqf_input_constructor
 from botorch.acquisition.monte_carlo import MCAcquisitionFunction
 from botorch.acquisition.objective import MCAcquisitionObjective

aepsych/acquisition/mutual_information.py

@@ -14,7 +14,7 @@
 
 import torch
 from aepsych.acquisition.monotonic_rejection import MonotonicMCAcquisition
-from aepsych.acquisition.objective import ProbitObjective
+from aepsych.objective import ProbitObjective
 from botorch.acquisition.input_constructors import acqf_input_constructor
 from botorch.acquisition.monte_carlo import MCAcquisitionFunction
 from botorch.acquisition.objective import MCAcquisitionObjective

aepsych/generators/monotonic_thompson_sampler_generator.py

@@ -9,7 +9,7 @@
 
 import numpy as np
 import torch
-from aepsych.acquisition.objective import ProbitObjective
+from aepsych.objective import ProbitObjective
 from aepsych.config import Config
 from aepsych.generators.base import AEPsychGenerator
 from aepsych.models.monotonic_rejection_gp import MonotonicRejectionGP

aepsych/generators/semi_p.py

@@ -8,7 +8,7 @@
 from typing import Type
 
 import torch
-from aepsych.acquisition.objective.semi_p import SemiPThresholdObjective
+from aepsych.objective.semi_p import SemiPThresholdObjective
 from aepsych.generators import OptimizeAcqfGenerator
 from aepsych.models.semi_p import SemiParametricGPModel
 

aepsych/likelihoods/semi_p.py

@@ -8,8 +8,8 @@
 from typing import Optional
 
 import torch
-from aepsych.acquisition.objective import AEPsychObjective, FloorProbitObjective
 from aepsych.config import Config
+from aepsych.objective import AEPsychObjective, FloorProbitObjective
 from gpytorch.likelihoods import _OneDimensionalLikelihood
 
 

aepsych/models/semi_p.py

@@ -13,9 +13,9 @@
 import gpytorch
 import numpy as np
 import torch
-from aepsych.acquisition.objective import FloorLogitObjective
+from aepsych.objective import FloorLogitObjective
 
-from aepsych.acquisition.objective.semi_p import SemiPThresholdObjective
+from aepsych.objective.semi_p import SemiPThresholdObjective
 from aepsych.config import Config
 from aepsych.likelihoods import BernoulliObjectiveLikelihood, LinearBernoulliLikelihood
 from aepsych.models import GPClassificationModel

aepsych/acquisition/objective/__init__.py → aepsych/objective/__init__.py

@@ -8,8 +8,9 @@
 import sys
 
 from ...config import Config
-from .objective import (
-    AEPsychObjective,
+from .aepsych_objective import AEPsychObjective
+from .probit import ProbitObjective
+from .floor_objective import (
     FloorGumbelObjective,
     FloorLogitObjective,
     FloorProbitObjective,

aepsych/objective/base.py

@@ -0,0 +1,13 @@
+# (c) Meta Platforms, Inc. and affiliates. Confidential and proprietary.
+
+# pyre-strict
+
+from typing import Optional
+
+from botorch.acquisition.objective import MCAcquisitionObjective
+from torch import Tensor
+
+
+class AEPsychObjective(MCAcquisitionObjective):
+    def inverse(self, samples: Tensor, X: Optional[Tensor] = None) -> Tensor:
+        raise NotImplementedError

aepsych/acquisition/objective/objective.py → aepsych/objective/floor_objective.py

@@ -5,52 +5,14 @@
 # This source code is licensed under the license found in the
 # LICENSE file in the root directory of this source tree.
 
-from __future__ import annotations
-
 from typing import Optional
 
 import torch
-from botorch.acquisition.objective import MCAcquisitionObjective
 from torch import Tensor
-from torch.distributions.normal import Normal
-
-
-class AEPsychObjective(MCAcquisitionObjective):
-    def inverse(self, samples: Tensor, X: Optional[Tensor] = None) -> Tensor:
-        raise NotImplementedError
-
-
-class ProbitObjective(AEPsychObjective):
-    """Probit objective
 
-    Transforms the input through the normal CDF (probit).
-    """
-
-    def forward(self, samples: Tensor, X: Optional[Tensor] = None) -> Tensor:
-        """Evaluates the objective (normal CDF).
-
-        Args:
-            samples (Tensor): GP samples.
-            X (Optional[Tensor], optional): ignored, here for compatibility
-                with MCAcquisitionObjective.
-
-        Returns:
-            Tensor: [description]
-        """
-        return Normal(loc=0, scale=1).cdf(samples.squeeze(-1))
-
-    def inverse(self, samples: Tensor, X: Optional[Tensor] = None) -> Tensor:
-        """Evaluates the inverse of the objective (normal PPF).
-
-        Args:
-            samples (Tensor): GP samples.
-            X (Optional[Tensor], optional): ignored, here for compatibility
-                with MCAcquisitionObjective.
+from torch.distributions.normal import Normal
 
-        Returns:
-            Tensor: [description]
-        """
-        return Normal(loc=0, scale=1).icdf(samples.squeeze(-1))
+from .base import AEPsychObjective
 
 
 class FloorLinkObjective(AEPsychObjective):

aepsych/objective/probit.py

@@ -0,0 +1,42 @@
+# (c) Meta Platforms, Inc. and affiliates. Confidential and proprietary.
+
+# pyre-strict
+
+from typing import Optional
+
+from torch import Tensor
+
+from .base import AEPsychObjective
+
+
+class ProbitObjective(AEPsychObjective):
+    """Probit objective
+
+    Transforms the input through the normal CDF (probit).
+    """
+
+    def forward(self, samples: Tensor, X: Optional[Tensor] = None) -> Tensor:
+        """Evaluates the objective (normal CDF).
+
+        Args:
+            samples (Tensor): GP samples.
+            X (Optional[Tensor], optional): ignored, here for compatibility
+                with MCAcquisitionObjective.
+
+        Returns:
+            Tensor: [description]
+        """
+        return Normal(loc=0, scale=1).cdf(samples.squeeze(-1))
+
+    def inverse(self, samples: Tensor, X: Optional[Tensor] = None) -> Tensor:
+        """Evaluates the inverse of the objective (normal PPF).
+
+        Args:
+            samples (Tensor): GP samples.
+            X (Optional[Tensor], optional): ignored, here for compatibility
+                with MCAcquisitionObjective.
+
+        Returns:
+            Tensor: [description]
+        """
+        return Normal(loc=0, scale=1).icdf(samples.squeeze(-1))

docs/gp_intro.md

@@ -144,7 +144,7 @@ By way of example, if our goal is to estimate the psychometric function at every
 
 From a technical perspective, there are two categories of acquisition functions in AEPsych:
 
-1. Monte Carlo (MC) acquisition functions. These acquisition functions can be evaluated by sampling from the model posterior. In the simplest case we sample from the latent perceptual field posterior, but we can also sample from other posteriors such as the response probability posterior, or response threshold posterior. The most common acquisition functions of this kind is BALD, or information maximization w.r.t. the response probability posterior (aka [`BernoulliMCMutualInformation`](https://aepsych.org/api/acquisition.html#aepsych.acquisition.mutual_information.BernoulliMCMutualInformation) in AEPsych), or BALV, or variance maximization (aka [`MCPosteriorVariance`](https://aepsych.org/api/acquisition.html#aepsych.acquisition.mc_posterior_variance.MCPosteriorVariance) in AEPsych). [Keeley et al. 2023](https://arxiv.org/abs/2302.01187) additionally introduced `ThresholdBALV`, which is BALV defined over the threshold rather than probability posterior for the semi-parametric GP models in AEPsych. To use `ThresholdBALV`, use `MCPosteriorVariance` with the `objective=[SemiPThresholdObjective](https://aepsych.org/api/acquisition.html#aepsych.acquisition.objective.SemiPThresholdObjective)` option. Because they only require sampling from the posterior, they can be used with any model which admits sampling from the posterior.
+1. Monte Carlo (MC) acquisition functions. These acquisition functions can be evaluated by sampling from the model posterior. In the simplest case we sample from the latent perceptual field posterior, but we can also sample from other posteriors such as the response probability posterior, or response threshold posterior. The most common acquisition functions of this kind is BALD, or information maximization w.r.t. the response probability posterior (aka [`BernoulliMCMutualInformation`](https://aepsych.org/api/acquisition.html#aepsych.acquisition.mutual_information.BernoulliMCMutualInformation) in AEPsych), or BALV, or variance maximization (aka [`MCPosteriorVariance`](https://aepsych.org/api/acquisition.html#aepsych.acquisition.mc_posterior_variance.MCPosteriorVariance) in AEPsych). [Keeley et al. 2023](https://arxiv.org/abs/2302.01187) additionally introduced `ThresholdBALV`, which is BALV defined over the threshold rather than probability posterior for the semi-parametric GP models in AEPsych. To use `ThresholdBALV`, use `MCPosteriorVariance` with the `objective=[SemiPThresholdObjective](https://aepsych.org/api/acquisition.html#aepsych.objective.SemiPThresholdObjective)` option. Because they only require sampling from the posterior, they can be used with any model which admits sampling from the posterior.
 2. Analytic acquisition functions. These acquisition functions require an explicit multivariate normal (MVN) posterior to be evaluated. They can often be faster to evaluate, and enable efficient computation of  _lookahead_ approaches, which compute the acquisition in expectation of the posterior as it will be given the next observation. As with MC acquisition functions, they can be evaluated w.r.t. the current or lookahead function posterior (as in [Zhao et al. 2021](https://proceedings.neurips.cc/paper_files/paper/2021/hash/50d2e70cdf7dd05be85e1b8df3f8ced4-Abstract.html), the lookahead level set (threshold) posterior, as described in [Letham et al. 2022](https://arxiv.org/abs/2203.09751), or any other posterior that can be approximated as MVN (e.g. the full semi-parametric posterior, with the MVN approximation taken after posterior inference.
 
 For more on analytic and MC acquisition functions, you can check the [BoTorch docs](https://botorch.org/docs/acquisition), and AEPsych supports all BoTorch acquisition functions, as well as a growing number of acquisition functions developed specifically for human psychophysics. From an application perspective, acquisition functions in AEPsych could be sorted based on their goals:

sphinx/source/acquisition.rst

@@ -1,4 +1,4 @@
-aepsych.acquisition 
+aepsych.acquisition
 ===========================
 
 Submodules
@@ -60,10 +60,10 @@ aepsych.acquisition.mutual\_information module
    :undoc-members:
    :show-inheritance:
 
-aepsych.acquisition.objective module
+aepsych.objective module
 ------------------------------------
 
-.. automodule:: aepsych.acquisition.objective
+.. automodule:: aepsych.objective
    :members:
    :undoc-members:
    :show-inheritance:

tests/acquisition/test_lse.py

@@ -10,7 +10,7 @@
 import numpy as np
 import torch
 from aepsych.acquisition.lse import MCLevelSetEstimation
-from aepsych.acquisition.objective import ProbitObjective
+from aepsych.objective import ProbitObjective
 from botorch.utils.testing import MockModel, MockPosterior
 from scipy.stats import norm
 

tests/acquisition/test_mi.py

@@ -13,7 +13,7 @@
     BernoulliMCMutualInformation,
     MonotonicBernoulliMCMutualInformation,
 )
-from aepsych.acquisition.objective import ProbitObjective
+from aepsych.objective import ProbitObjective
 from aepsych.benchmark.test_functions import f_1d
 from aepsych.generators import (
     MonotonicRejectionGenerator,

tests/acquisition/test_monotonic.py

@@ -7,7 +7,7 @@
 
 import torch
 from aepsych.acquisition.monotonic_rejection import MonotonicMCLSE
-from aepsych.acquisition.objective import ProbitObjective
+from aepsych.objective import ProbitObjective
 from aepsych.models.derivative_gp import MixedDerivativeVariationalGP
 from botorch.acquisition.objective import IdentityMCObjective
 from botorch.utils.testing import BotorchTestCase

tests/acquisition/test_objective.py

@@ -10,7 +10,7 @@
 
 import numpy as np
 import torch
-from aepsych.acquisition.objective import (
+from aepsych.objective import (
     FloorGumbelObjective,
     FloorLogitObjective,
     FloorProbitObjective,

tests/models/test_monotonic_rejection_gp.py

@@ -15,7 +15,7 @@
 
 
 from aepsych.acquisition.monotonic_rejection import MonotonicMCLSE
-from aepsych.acquisition.objective import ProbitObjective
+from aepsych.objective import ProbitObjective
 from aepsych.generators import MonotonicRejectionGenerator
 from aepsych.models import MonotonicRejectionGP
 from aepsych.strategy import Strategy

tests/models/test_pairwise_probit.py

@@ -13,7 +13,7 @@
 import numpy.testing as npt
 import torch
 from aepsych import server, utils_logging
-from aepsych.acquisition.objective import ProbitObjective
+from aepsych.objective import ProbitObjective
 from aepsych.benchmark.test_functions import f_1d, f_2d, f_pairwise, new_novel_det
 from aepsych.config import Config
 from aepsych.generators import OptimizeAcqfGenerator, SobolGenerator

tests/models/test_semi_p.py

@@ -13,7 +13,7 @@
 import torch
 from aepsych.acquisition import MCPosteriorVariance
 from aepsych.acquisition.lookahead import GlobalMI
-from aepsych.acquisition.objective import (
+from aepsych.objective import (
     FloorGumbelObjective,
     FloorLogitObjective,
     FloorProbitObjective,
@@ -22,7 +22,7 @@
 from aepsych.generators import OptimizeAcqfGenerator, SobolGenerator
 from aepsych.likelihoods import BernoulliObjectiveLikelihood
 from aepsych.strategy import SequentialStrategy, Strategy
-from aepsych.acquisition.objective.semi_p import (
+from aepsych.objective.semi_p import (
     SemiPProbabilityObjective,
     SemiPThresholdObjective,
 )

tests/test_config.py

@@ -13,7 +13,7 @@
 import torch
 from aepsych.acquisition import EAVC, MCLevelSetEstimation
 from aepsych.acquisition.monotonic_rejection import MonotonicMCLSE
-from aepsych.acquisition.objective import FloorGumbelObjective, ProbitObjective
+from aepsych.objective import FloorGumbelObjective, ProbitObjective
 from aepsych.config import Config
 from aepsych.generators import (
     MonotonicRejectionGenerator,

website/_tutorials/Semi_P_tutorial.html

@@ -59,7 +59,7 @@ <h3 id="Semi-parametric-psychophysical-model-tutorial">Semi-parametric psychophy
 <div class="highlight hl-ipython3"><pre><span></span><span class="c1">### SemiP imports</span>
 
 <span class="kn">from</span> <span class="nn">aepsych.likelihoods</span> <span class="kn">import</span> <span class="n">BernoulliObjectiveLikelihood</span>
-<span class="kn">from</span> <span class="nn">aepsych.acquisition.objective</span> <span class="kn">import</span> <span class="p">(</span>
+<span class="kn">from</span> <span class="nn">aepsych.objective</span> <span class="kn">import</span> <span class="p">(</span>
     <span class="n">FloorGumbelObjective</span><span class="p">,</span>
     <span class="n">FloorLogitObjective</span><span class="p">,</span>
     <span class="n">FloorProbitObjective</span><span class="p">,</span>
@@ -85,7 +85,7 @@ <h4 id="Test-function">Test function<a class="anchor-link" href="#Test-function"
 <div class="inner_cell">
 <div class="input_area">
 <div class="highlight hl-ipython3"><pre><span></span><span class="c1"># Detection test function</span>
-<span class="n">lb</span> <span class="o">=</span> <span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">]</span> 
+<span class="n">lb</span> <span class="o">=</span> <span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">]</span>
 <span class="n">ub</span> <span class="o">=</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">]</span>
 
 <span class="n">xgrid</span> <span class="o">=</span> <span class="n">dim_grid</span><span class="p">(</span><span class="n">lower</span><span class="o">=</span><span class="n">lb</span><span class="p">,</span> <span class="n">upper</span><span class="o">=</span><span class="n">ub</span><span class="p">,</span> <span class="n">dim</span><span class="o">=</span><span class="mi">2</span><span class="p">)</span>
@@ -149,7 +149,7 @@ <h4 id="Fitting-the-model">Fitting the model<a class="anchor-link" href="#Fittin
 <div class="inner_cell">
 <div class="input_area">
 <div class="highlight hl-ipython3"><pre><span></span><span class="c1">### fit SemiP models</span>
-<span class="n">semip_model</span> <span class="o">=</span> <span class="n">SemiParametricGPModel</span><span class="p">(</span><span class="n">lb</span><span class="o">=</span><span class="n">lb</span><span class="p">,</span> <span class="n">ub</span><span class="o">=</span><span class="n">ub</span><span class="p">,</span> <span class="n">dim</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span> <span class="n">stim_dim</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span> 
+<span class="n">semip_model</span> <span class="o">=</span> <span class="n">SemiParametricGPModel</span><span class="p">(</span><span class="n">lb</span><span class="o">=</span><span class="n">lb</span><span class="p">,</span> <span class="n">ub</span><span class="o">=</span><span class="n">ub</span><span class="p">,</span> <span class="n">dim</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span> <span class="n">stim_dim</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
                                     <span class="n">likelihood</span><span class="o">=</span><span class="n">LinearBernoulliLikelihood</span><span class="p">(</span><span class="n">objective</span><span class="o">=</span><span class="n">FloorProbitObjective</span><span class="p">(</span><span class="n">floor</span><span class="o">=</span><span class="mi">0</span><span class="p">)))</span>
 <span class="n">approx_model</span> <span class="o">=</span> <span class="n">HadamardSemiPModel</span><span class="p">(</span><span class="n">lb</span><span class="o">=</span><span class="n">lb</span><span class="p">,</span> <span class="n">ub</span><span class="o">=</span><span class="n">ub</span><span class="p">,</span> <span class="n">dim</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span> <span class="n">stim_dim</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
                                   <span class="n">likelihood</span><span class="o">=</span><span class="n">BernoulliObjectiveLikelihood</span><span class="p">(</span><span class="n">objective</span><span class="o">=</span><span class="n">FloorProbitObjective</span><span class="p">(</span><span class="n">floor</span><span class="o">=</span><span class="mi">0</span><span class="p">)))</span>
@@ -332,4 +332,4 @@ <h1 id="Active-learning">Active learning<a class="anchor-link" href="#Active-lea
 </div>
 </div>
 </div>
-</div>
+</div>

website/pages/api/_modules/aepsych/acquisition/lse.html

@@ -20,7 +20,7 @@ <h1>Source code for aepsych.acquisition.lse</h1><div class="highlight"><pre>
 <span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">Optional</span><span class="p">,</span> <span class="n">Union</span>
 
 <span class="kn">import</span> <span class="nn">torch</span>
-<span class="kn">from</span> <span class="nn">aepsych.acquisition.objective</span> <span class="kn">import</span> <span class="n">ProbitObjective</span>
+<span class="kn">from</span> <span class="nn">aepsych.objective</span> <span class="kn">import</span> <span class="n">ProbitObjective</span>
 <span class="kn">from</span> <span class="nn">botorch.acquisition.input_constructors</span> <span class="kn">import</span> <span class="n">acqf_input_constructor</span>
 <span class="kn">from</span> <span class="nn">botorch.acquisition.monte_carlo</span> <span class="kn">import</span> <span class="p">(</span>
     <span class="n">MCAcquisitionFunction</span><span class="p">,</span>
@@ -165,4 +165,4 @@ <h3 id="searchlabel">Quick search</h3>
 </div>
 </div>
 <div class="clearer"></div>
-</div></div>
+</div></div>