trieste.data

This module contains utilities for Observer data.

Module Contents

class Dataset

Container for the query points and corresponding observations from an Observer.

query_points: trieste.types.TensorType

The points at which the Observer was queried.

observations: trieste.types.TensorType

The observed output of the Observer for each query point.

__post_init__() → None

Raises:

ValueError (or InvalidArgumentError) – If query_points or observations have rank less than two, or they have unequal shape in any but their last dimension.

__add__(rhs: Dataset) → Dataset

Return the Dataset whose query points are the result of concatenating the query_points in each Dataset along the zeroth axis, and the same for the observations. For example:

>>> d1 = Dataset(
...     tf.constant([[0.1, 0.2], [0.3, 0.4]]),
...     tf.constant([[0.5, 0.6], [0.7, 0.8]])
... )
>>> d2 = Dataset(tf.constant([[0.9, 1.0]]), tf.constant([[1.1, 1.2]]))
>>> (d1 + d2).query_points
<tf.Tensor: shape=(3, 2), dtype=float32, numpy=
array([[0.1, 0.2],
       [0.3, 0.4],
       [0.9, 1. ]], dtype=float32)>
>>> (d1 + d2).observations
<tf.Tensor: shape=(3, 2), dtype=float32, numpy=
array([[0.5, 0.6],
       [0.7, 0.8],
       [1.1, 1.2]], dtype=float32)>

Parameters:

rhs – A Dataset with the same shapes as this one, except in the zeroth dimension, which can have any size.

Returns:

The result of concatenating the Datasets.

Raises:

InvalidArgumentError – If the shapes of the query_points in each Dataset differ in any but the zeroth dimension. The same applies for observations.

__len__() → tensorflow.Tensor

Returns:

The number of query points, or equivalently the number of observations.

astuple() → tuple[trieste.types.TensorType, trieste.types.TensorType]

Note: Unlike the standard library function dataclasses.astuple, this method does not deepcopy the attributes.

Returns:

A 2-tuple of the query_points and observations.

check_and_extract_fidelity_query_points(query_points: trieste.types.TensorType, max_fidelity: int | None = None) → tuple[trieste.types.TensorType, trieste.types.TensorType]

Check whether the final column of a tensor is close enough to ints to be reasonably considered to represent fidelities.

The final input column of multi-fidelity data should be a reference to the fidelity of the query point. We cannot have mixed type tensors, but we can check that thhe final column values are suitably close to integers.

Parameters:

query_points – Data to check final column of.

Raise:

ValueError: If there are not enough columns to be multifidelity data

Raises:

InvalidArgumentError – If any value in the final column is far from an integer

Returns:

Query points without fidelity column and the fidelities of each of the query points

split_dataset_by_fidelity(dataset: Dataset, num_fidelities: int) → Sequence[Dataset]

Split dataset into individual datasets without fidelity information

Parameters:

• dataset – Dataset for which to split fidelities

• num_fidlities – Number of fidelities in the problem (not just dataset)

Returns:

Ordered list of datasets with lowest fidelity at index 0 and highest at -1

get_dataset_for_fidelity(dataset: Dataset, fidelity: int) → Dataset

Get a dataset with only the specified fidelity of data in

Parameters:

• dataset – The dataset from which to extract the single fidelity data

• fidelity – The fidelity to extract the data for

Returns:

Dataset with a single fidelity and no fidelity column

add_fidelity_column(query_points: trieste.types.TensorType, fidelity: int) → trieste.types.TensorType

Add fidelity column to query_points without fidelity data

Parameters:

• query_points – query points without fidelity to add fidelity column to

• fidelity – fidelity to populate fidelity column with

Returns:

TensorType of query points with fidelity column added