Preprocessing Quantize Scalar

Source header: cuvs/preprocessing/quantize/scalar.h

C API for Scalar Quantizer

cuvsScalarQuantizerParams

Scalar quantizer parameters.

1 struct cuvsScalarQuantizerParams { ... };

cuvsScalarQuantizerParamsCreate

Allocate Scalar Quantizer params, and populate with default values

1 CUVS_EXPORT cuvsError_t cuvsScalarQuantizerParamsCreate(cuvsScalarQuantizerParams_t* params);

Parameters

Name	Direction	Type	Description
`params`	in	`cuvsScalarQuantizerParams_t*`	cuvsScalarQuantizerParams_t to allocate

Returns

CUVS_EXPORT cuvsError_t

cuvsScalarQuantizerParamsDestroy

De-allocate Scalar Quantizer params

1 CUVS_EXPORT cuvsError_t cuvsScalarQuantizerParamsDestroy(cuvsScalarQuantizerParams_t params);

Parameters

Name	Direction	Type	Description
`params`	in	`cuvsScalarQuantizerParams_t`

Returns

CUVS_EXPORT cuvsError_t

cuvsScalarQuantizer

Defines and stores scalar for quantisation upon training

The quantization is performed by a linear mapping of an interval in the float data type to the full range of the quantized int type.

1 typedef struct { ... } cuvsScalarQuantizer;

Fields

Name	Type	Description
`min_`	`double`
`max_`	`double`

cuvsScalarQuantizerCreate

Allocate Scalar Quantizer and populate with default values

1 CUVS_EXPORT cuvsError_t cuvsScalarQuantizerCreate(cuvsScalarQuantizer_t* quantizer);

Parameters

Name	Direction	Type	Description
`quantizer`	in	`cuvsScalarQuantizer_t*`	cuvsScalarQuantizer_t to allocate

Returns

CUVS_EXPORT cuvsError_t

cuvsScalarQuantizerDestroy

De-allocate Scalar Quantizer

1 CUVS_EXPORT cuvsError_t cuvsScalarQuantizerDestroy(cuvsScalarQuantizer_t quantizer);

Parameters

Name	Direction	Type	Description
`quantizer`	in	`cuvsScalarQuantizer_t`

Returns

CUVS_EXPORT cuvsError_t

cuvsScalarQuantizerTrain

Trains a scalar quantizer to be used later for quantizing the dataset.

1 CUVS_EXPORT cuvsError_t cuvsScalarQuantizerTrain(cuvsResources_t res,
2 cuvsScalarQuantizerParams_t params,
3 DLManagedTensor* dataset,
4 cuvsScalarQuantizer_t quantizer);

Parameters

Name	Direction	Type	Description
`res`	in	`cuvsResources_t`	raft resource
`params`	in	`cuvsScalarQuantizerParams_t`	configure scalar quantizer, e.g. quantile
`dataset`	in	`DLManagedTensor*`	a row-major host or device matrix
`quantizer`	out	`cuvsScalarQuantizer_t`	trained scalar quantizer

Returns

CUVS_EXPORT cuvsError_t

cuvsScalarQuantizerTransform

Applies quantization transform to given dataset

1 CUVS_EXPORT cuvsError_t cuvsScalarQuantizerTransform(cuvsResources_t res,
2 cuvsScalarQuantizer_t quantizer,
3 DLManagedTensor* dataset,
4 DLManagedTensor* out);

Parameters

Name	Direction	Type	Description
`res`	in	`cuvsResources_t`	raft resource
`quantizer`	in	`cuvsScalarQuantizer_t`	a scalar quantizer
`dataset`	in	`DLManagedTensor*`	a row-major host or device matrix to transform
`out`	out	`DLManagedTensor*`	a row-major host or device matrix to store transformed data

Returns

CUVS_EXPORT cuvsError_t

cuvsScalarQuantizerInverseTransform

Perform inverse quantization step on previously quantized dataset

1 CUVS_EXPORT cuvsError_t cuvsScalarQuantizerInverseTransform(cuvsResources_t res,
2 cuvsScalarQuantizer_t quantizer,
3 DLManagedTensor* dataset,
4 DLManagedTensor* out);

Note that depending on the chosen data types train dataset the conversion is not lossless.

Parameters

Name	Direction	Type	Description
`res`	in	`cuvsResources_t`	raft resource
`quantizer`	in	`cuvsScalarQuantizer_t`	a scalar quantizer
`dataset`	in	`DLManagedTensor*`	a row-major host or device matrix
`out`	out	`DLManagedTensor*`	a row-major host or device matrix

Returns

CUVS_EXPORT cuvsError_t