tfp.experimental.bayesopt.acquisition.ParallelProbabilityOfImprovement

Parallel probability of improvement acquisition function.

Inherits From: AcquisitionFunction

tfp.experimental.bayesopt.acquisition.ParallelProbabilityOfImprovement(
    predictive_distribution,
    observations,
    seed=None,
    num_samples=100,
    transform_fn=None
)

Computes the q-PI from a multivariate observation model. This is also known as batch probability of improvement.

Requires that predictive_distribution has a sample method.

Examples

Build and evaluate a Parallel Probability of Improvement acquisition function.

import numpy as np
import tensorflow_probability as tfp

tfd = tfp.distributions
tfpk = tfp.math.psd_kernels
tfp_acq = tfp.experimental.bayesopt.acquisition

# Sample 10 20-dimensional index points and associated observations.
index_points = np.random.uniform(size=[10, 20])
observations = np.random.uniform(size=[10])

# Build a Student T Process regression model conditioned on observed data.
dist = tfd.StudentTProcessRegressionModel(
    kernel=tfpk.ExponentiatedQuadratic(),
    df=5.,
    observation_index_points=index_points,
    observations=observations)

# Define a Parallel Probability of Improvement acquisition function.
stp_pei = tfp_acq.ParallelProbabilityOfImprovement(
    predictive_distribution=dist,
    observations=observations,
    num_samples=10_000)

# Evaluate the acquisition function at a new set of index points.
pred_index_points = np.random.uniform(size=[6, 20])
acq_fn_vals = stp_pei(pred_index_points)  # Has shape [6].

predictive_distribution tfd.Distribution-like, the distribution over observations at a set of index points. Must have a sample method.
observations Float Tensor of observations. Shape has the form [b1, ..., bB, e], where e is the number of index points (such that the event shape of predictive_distribution is [e]) and [b1, ..., bB] is broadcastable with the batch shape of predictive_distribution.
seed PRNG seed; see tfp.random.sanitize_seed for details.
num_samples The number of samples to use for the Parallel Probability of Improvement approximation.
transform_fn Optional Python Callable that transforms objective values. This is used for optimizing a composite grey box function g(f(x)) where f is our black box function and g is transform_fn.

is_parallel Python bool indicating whether the acquisition function is parallel.

Parallel (batched) acquisition functions evaluate batches of points rather than single points.

num_samples

observations Float Tensor of observations.
predictive_distribution The distribution over observations at a set of index points.
seed PRNG seed.
transform_fn

Methods

__call__

View source

__call__(
    **kwargs
)

Computes the Parallel Probability of Improvement.

Args
**kwargs Keyword args passed on to the sample method of predictive_distribution.

Returns
Parallel Probability of improvement at index points implied by predictive_distribution (or overridden in **kwargs).