eulerpi.core.data_transformations module

Data transformations can be used to improve the performance of the inference function by improving the quality of the kernel density estimate.

This subpackage contains all predefined data transformations and an abstract base class for custom data transformations.

class AffineTransformation(A: Array, b: Array)

Bases: DataTransformation

Class for applying an affine data transformation, y=Ax+b

jacobian(data: Array) → Array

Calculate the jacobian of the transformation at the given data point(s).

Parameters:

data (jnp.ndarray) – The data at which the jacobian should be evaluated.

Raises:

NotImplementedError – Raised if the jacobian is not implemented in the subclass.

Returns:

The jacobian of the transformation at the given data point(s).

Return type:

jnp.ndarray

transform(data: float64 | Array) → float64 | Array

Normalize the given data.

Parameters:

data (Union[jnp.double, jnp.ndarray]) – The data to be normalized. Columns correspond to different dimensions. Rows correspond to different observations.

Returns:

The normalized data.

Return type:

Union[jnp.double, jnp.ndarray]

class DataIdentity

Bases: DataTransformation

The identity transformation. Does not change the data.

jacobian(data: Array) → Array

Returns the identity matrix.

Parameters:

data (jnp.ndarray) – The data at which the jacobian should be evaluated.

Returns:

The identity matrix.

Return type:

jnp.ndarray

transform(data: Array) → Array

Returns the data unchanged.

Parameters:

data (jnp.ndarray) – The data which should be transformed.

Returns:

The data unchanged.

Return type:

jnp.ndarray

class DataNormalization(data: Array)

Bases: AffineTransformation

Class for normalizing data. The data is normalized by subtracting the mean and multiplying by the inverse of the Cholesky decomposition of the covariance matrix.

class DataPCA(data: Array, n_components: int | str)

Bases: DataTransformation

The DataPCA class can be used to transform the data using the Principal Component Analysis.

jacobian(data: Array) → Array

Return the jacobian of the pca transformation at the given data point(s).

Parameters:

data (jnp.ndarray) – The data point(s) at which the jacobian should be evaluated.

Returns:

The jacobian of the pca transformation at the given data point(s).

Return type:

jnp.ndarray

transform(data: Array) → Array

Transform the given data using the PCA.

Parameters:

data (jnp.ndarray) – The data to be transformed.

Returns:

The data projected onto and expressed in the basis of the principal components.

Return type:

jnp.ndarray

class DataTransformation

Bases: ABC

Abstract base class for all data transformations

Data transformations can be used to improve the performance of the inference function by improving the quality of the kernel density estimate.

abstract jacobian(data: Array) → Array

Calculate the jacobian of the transformation at the given data point(s).

Parameters:

data (jnp.ndarray) – The data at which the jacobian should be evaluated.

Raises:

NotImplementedError – Raised if the jacobian is not implemented in the subclass.

Returns:

The jacobian of the transformation at the given data point(s).

Return type:

jnp.ndarray

abstract transform(data: Array) → Array

Transform the given data point(s)

Parameters:

data (jnp.ndarray) – The data to be transformed. Columns correspond to different dimensions. Rows correspond to different observations.

Raises:

NotImplementedError – Raised if the transform is not implemented in the subclass.

Returns:

The transformed data point(s).

Return type:

jnp.ndarray