PCA

Source header: cuvs/preprocessing/pca.h

C API for PCA (Principal Component Analysis)

cuvsPcaSolver

Solver algorithm for PCA eigen decomposition.

1 enum cuvsPcaSolver {
2   CUVS_PCA_COV_EIG_DQ = 0,
3   CUVS_PCA_COV_EIG_JACOBI = 1
4 };

Values

Name	Value
`CUVS_PCA_COV_EIG_DQ`	`0`
`CUVS_PCA_COV_EIG_JACOBI`	`1`

cuvsPcaParams

Parameters for PCA decomposition.

1 struct cuvsPcaParams {
2   int n_components;
3   bool copy;
4   bool whiten;
5   enum cuvsPcaSolver algorithm;
6   float tol;
7   int n_iterations;
8 };

Fields

Name	Type	Description
`n_components`	`int`	Number of principal components to keep.
`copy`	`bool`	If false, data passed to fit are overwritten and running fit(X).transform(X) will not yield the expected results; use fit_transform(X) instead.
`whiten`	`bool`	When true the component vectors are multiplied by the square root of n_samples and then divided by the singular values to ensure uncorrelated outputs with unit component-wise variances.
`algorithm`	`enum cuvsPcaSolver`	Solver algorithm to use.
`tol`	`float`	Tolerance for singular values (used by Jacobi solver).
`n_iterations`	`int`	Number of iterations for the power method (Jacobi solver).

cuvsPcaParamsCreate

Allocate PCA params and populate with default values.

1 cuvsError_t cuvsPcaParamsCreate(cuvsPcaParams_t* params);

Parameters

Name	Direction	Type	Description
`params`	out	`cuvsPcaParams_t*`	cuvsPcaParams_t to allocate

Returns

cuvsError_t

cuvsPcaParamsDestroy

De-allocate PCA params.

1 cuvsError_t cuvsPcaParamsDestroy(cuvsPcaParams_t params);

Parameters

Name	Direction	Type	Description
`params`	in	`cuvsPcaParams_t`	cuvsPcaParams_t to de-allocate

Returns

cuvsError_t

cuvsPcaFit

Perform PCA fit operation.

1 cuvsError_t cuvsPcaFit(cuvsResources_t res,
2 cuvsPcaParams_t params,
3 DLManagedTensor* input,
4 DLManagedTensor* components,
5 DLManagedTensor* explained_var,
6 DLManagedTensor* explained_var_ratio,
7 DLManagedTensor* singular_vals,
8 DLManagedTensor* mu,
9 DLManagedTensor* noise_vars,
10 bool flip_signs_based_on_U);

Computes the principal components, explained variances, singular values, and column means from the input data.

Parameters

Name	Direction	Type	Description
`res`	in	`cuvsResources_t`	cuvsResources_t opaque C handle
`params`	in	`cuvsPcaParams_t`	PCA parameters
`input`	inout	`DLManagedTensor*`	input data [n_rows x n_cols] (col-major, float32, device)
`components`	out	`DLManagedTensor*`	principal components [n_components x n_cols] (col-major, float32, device)
`explained_var`	out	`DLManagedTensor*`	explained variances [n_components] (float32, device)
`explained_var_ratio`	out	`DLManagedTensor*`	explained variance ratios [n_components] (float32, device)
`singular_vals`	out	`DLManagedTensor*`	singular values [n_components] (float32, device)
`mu`	out	`DLManagedTensor*`	column means [n_cols] (float32, device)
`noise_vars`	out	`DLManagedTensor*`	noise variance [1] (float32, device)
`flip_signs_based_on_U`	in	`bool`	whether to determine signs by U (true) or V.T (false)

Returns

cuvsError_t

cuvsPcaFitTransform

Perform PCA fit and transform in a single operation.

1 cuvsError_t cuvsPcaFitTransform(cuvsResources_t res,
2 cuvsPcaParams_t params,
3 DLManagedTensor* input,
4 DLManagedTensor* trans_input,
5 DLManagedTensor* components,
6 DLManagedTensor* explained_var,
7 DLManagedTensor* explained_var_ratio,
8 DLManagedTensor* singular_vals,
9 DLManagedTensor* mu,
10 DLManagedTensor* noise_vars,
11 bool flip_signs_based_on_U);

Computes the principal components and transforms the input data into the eigenspace.

Parameters

Name	Direction	Type	Description
`res`	in	`cuvsResources_t`	cuvsResources_t opaque C handle
`params`	in	`cuvsPcaParams_t`	PCA parameters
`input`	inout	`DLManagedTensor*`	input data [n_rows x n_cols] (col-major, float32, device)
`trans_input`	out	`DLManagedTensor*`	transformed data [n_rows x n_components] (col-major, float32, device)
`components`	out	`DLManagedTensor*`	principal components [n_components x n_cols] (col-major, float32, device)
`explained_var`	out	`DLManagedTensor*`	explained variances [n_components] (float32, device)
`explained_var_ratio`	out	`DLManagedTensor*`	explained variance ratios [n_components] (float32, device)
`singular_vals`	out	`DLManagedTensor*`	singular values [n_components] (float32, device)
`mu`	out	`DLManagedTensor*`	column means [n_cols] (float32, device)
`noise_vars`	out	`DLManagedTensor*`	noise variance [1] (float32, device)
`flip_signs_based_on_U`	in	`bool`	whether to determine signs by U (true) or V.T (false)

Returns

cuvsError_t

cuvsPcaTransform

Perform PCA transform operation.

1 cuvsError_t cuvsPcaTransform(cuvsResources_t res,
2 cuvsPcaParams_t params,
3 DLManagedTensor* input,
4 DLManagedTensor* components,
5 DLManagedTensor* singular_vals,
6 DLManagedTensor* mu,
7 DLManagedTensor* trans_input);

Transforms the input data into the eigenspace using previously computed principal components.

Parameters

Name	Direction	Type	Description
`res`	in	`cuvsResources_t`	cuvsResources_t opaque C handle
`params`	in	`cuvsPcaParams_t`	PCA parameters
`input`	inout	`DLManagedTensor*`	data to transform [n_rows x n_cols] (col-major, float32, device)
`components`	in	`DLManagedTensor*`	principal components [n_components x n_cols] (col-major, float32, device)
`singular_vals`	in	`DLManagedTensor*`	singular values [n_components] (float32, device)
`mu`	in	`DLManagedTensor*`	column means [n_cols] (float32, device)
`trans_input`	out	`DLManagedTensor*`	transformed data [n_rows x n_components] (col-major, float32, device)

Returns

cuvsError_t

cuvsPcaInverseTransform

Perform PCA inverse transform operation.

1 cuvsError_t cuvsPcaInverseTransform(cuvsResources_t res,
2 cuvsPcaParams_t params,
3 DLManagedTensor* trans_input,
4 DLManagedTensor* components,
5 DLManagedTensor* singular_vals,
6 DLManagedTensor* mu,
7 DLManagedTensor* output);

Transforms data from the eigenspace back to the original space.

Parameters

Name	Direction	Type	Description
`res`	in	`cuvsResources_t`	cuvsResources_t opaque C handle
`params`	in	`cuvsPcaParams_t`	PCA parameters
`trans_input`	in	`DLManagedTensor*`	transformed data [n_rows x n_components] (col-major, float32, device)
`components`	in	`DLManagedTensor*`	principal components [n_components x n_cols] (col-major, float32, device)
`singular_vals`	in	`DLManagedTensor*`	singular values [n_components] (float32, device)
`mu`	in	`DLManagedTensor*`	column means [n_cols] (float32, device)
`output`	out	`DLManagedTensor*`	reconstructed data [n_rows x n_cols] (col-major, float32, device)

Returns

cuvsError_t

1	enum cuvsPcaSolver {
2	CUVS_PCA_COV_EIG_DQ = 0,
3	CUVS_PCA_COV_EIG_JACOBI = 1
4	};

1	struct cuvsPcaParams {
2	int n_components;
3	bool copy;
4	bool whiten;
5	enum cuvsPcaSolver algorithm;
6	float tol;
7	int n_iterations;
8	};

1	cuvsError_t cuvsPcaFit(cuvsResources_t res,
2	cuvsPcaParams_t params,
3	DLManagedTensor* input,
4	DLManagedTensor* components,
5	DLManagedTensor* explained_var,
6	DLManagedTensor* explained_var_ratio,
7	DLManagedTensor* singular_vals,
8	DLManagedTensor* mu,
9	DLManagedTensor* noise_vars,
10	bool flip_signs_based_on_U);

1	cuvsError_t cuvsPcaFitTransform(cuvsResources_t res,
2	cuvsPcaParams_t params,
3	DLManagedTensor* input,
4	DLManagedTensor* trans_input,
5	DLManagedTensor* components,
6	DLManagedTensor* explained_var,
7	DLManagedTensor* explained_var_ratio,
8	DLManagedTensor* singular_vals,
9	DLManagedTensor* mu,
10	DLManagedTensor* noise_vars,
11	bool flip_signs_based_on_U);

1	cuvsError_t cuvsPcaTransform(cuvsResources_t res,
2	cuvsPcaParams_t params,
3	DLManagedTensor* input,
4	DLManagedTensor* components,
5	DLManagedTensor* singular_vals,
6	DLManagedTensor* mu,
7	DLManagedTensor* trans_input);

1	cuvsError_t cuvsPcaInverseTransform(cuvsResources_t res,
2	cuvsPcaParams_t params,
3	DLManagedTensor* trans_input,
4	DLManagedTensor* components,
5	DLManagedTensor* singular_vals,
6	DLManagedTensor* mu,
7	DLManagedTensor* output);