Tutorial

Sampler

Samplers are used to create simple sample sets. The method sample_points() takes a set of arguments and returns a numpy array. Each sampler will have its own implementation, and so will take a unique set of arguments.

Example

To create a set of 10 points using the latin hypercube strategy in the unit square:

from trata.sampler import LatinHyperCubeSampler
points = LatinHyperCubeSampler.sample_points(num_points=10, box=[[0, 1], [0, 1]])

Full description of samplers and their arguments can be found in this notebook.

Composite Samples

A Composite Samples object makes it easier to generate more complicated sample sets. With it, you can specify a set of variables and sample using those variables multiple times.

A full tutorial of the Sampling component can be found in this notebook.

Examples

Two Continuous Variables

We want to sample over 2 continuous variables, X and Y. A typical sampling strategy is to look at the corners, the faces, and some points in the interior. We’ll let X have a range [0.1, 0.2] and Y have a range [1.5, 2.5]. We must also decide on a default value for each variable. We’ll let the default be the midpoint of the range.

To start, we must create a composite_samples object to hold our sample points.

samples = composite_samples.Samples()

Next, we can define our two variables in samples.

samples.set_continuous_variable('X', 0.1, 0.15, 0.2)
samples.set_continuous_variable('Y', 1.5, 2.0, 2.5)

Now that the variables are defined in samples we can start generating our sample set. Each sampler implements a sampling strategy. We need 3 samplers to generate our sample set: CornerSampler, OneAtATimeSampler, and LatinHyperCubeSampler We’ll instantiate each of these so we can use them later.

corner_sampler = trata.CornerSampler()
face_sampler = trata.OneAtATimeSampler()
lhs_sampler = trata.LatinHyperCubeSampler()

To actually generate the sample points we use the generate_samples method. This method takes a list of variable names and a sampler (with a set of keyword arguments). It calls the sampler’s sample_points method using the keyword arguments and stores those points in the samples object under the specified variables.

samples.generate_samples(['X', 'Y'], corner_sampler)
samples.generate_samples(['X', 'Y'], face_sampler, do_oat=True)
samples.generate_samples(['X', 'Y'], lhs_sampler, num_points=10)

Now our sample set is stored in the samples object. To access these points we’ll use get_points. This method simply returns all the points that have been generated.

sample_points = samples.get_points()