data module

This section contains the Python API reference for the uncertainty_toolbox.data module, which contains code for importing and generating data.

uncertainty_toolbox.data Module

Code for importing and generating data.

uncertainty_toolbox.data.synthetic_arange_random(num_points=10)

Dataset of evenly spaced points and identity function (with some randomization).

This function returns predictions and predictive uncertainties (given as standard deviations) from some hypothetical uncertainty model, along with true input x and output y data points.

Parameters

num_points (int) – The number of data points in the set.

Return type

Tuple[ndarray, ndarray, ndarray, ndarray]

Returns

  • The y predictions given by a hypothetical predictive uncertainty model. These are the true values of y but with uniform noise added.

  • The standard deviations given by a hypothetical predictive uncertainty model. These are the errors between the predictions and the truth plus some unifom noise.

  • The true outputs y.

  • The true inputs x.

uncertainty_toolbox.data.synthetic_sine_heteroscedastic(n_points=10)

Return samples from “synthetic sine” heteroscedastic noisy function.

This returns a synthetic dataset which can be used to train and assess a predictive uncertainty model.

Parameters

n_points (int) – The number of data points in the set.

Return type

Tuple[ndarray, ndarray, ndarray, ndarray]

Returns

  • Predicted output points y.

  • Predictive uncertainties, defined using standard deviation of added noise.

  • True output points y.

  • True input points x.