All Products
Search
Document Center

:Proxima Cluster parameters

Last Updated:Mar 18, 2026

1. Clustering

1.1 KmeansCluster / BatchKmeansCluster

Parameter

Type

Default value

Description

proxima.general.cluster.count

UINT32

0

Number of centroids.

proxima.kmeans.cluster.count

UINT32

0

Number of centroids. This parameter overrides the general parameter but is overridden by the suggested K value.

proxima.kmeans.cluster.shard_factor

FLOAT

16.0f

Tuning factor for multi-threaded concurrency.

proxima.kmeans.cluster.epsilon

DOUBLE

FL_EPSILON

Clustering convergence precision.

proxima.kmeans.cluster.max_iterations

UINT32

20

Maximum number of iterations.

proxima.kmeans.cluster.purge_empty

BOOL

false

Specifies whether to delete empty centroids.

proxima.kmeans.cluster.seeker_class

STRING

LinearSeeker

The algorithm class for finding centroids.

proxima.kmeans.cluster.seeker_params

IndexParams

Parameters for the centroid finding algorithm class.

IndexParams object

1.2 GpuKmeansCluster

Parameter name

Type

Default value

Note:

proxima.general.cluster.count

UINT32

0

Number of centroids.

proxima.kmeans.cluster.count

UINT32

0

Number of centroids. This parameter overrides the general parameter but is overridden by the suggested K value.

proxima.kmeans.cluster.epsilon

DOUBLE

FL_EPSILON

Clustering convergence precision.

proxima.kmeans.cluster.max_iterations

UINT32

100

Maximum number of iterations.

proxima.kmeans.cluster.purge_empty

BOOL

false

Specifies whether to delete empty centroids.

1.3 MiniBatchKmeansCluster

Parameter name

Type

Default value

Note:

proxima.general.cluster.count

UINT32

0

Number of centroids.

proxima.minibatchkmeans.cluster.count

UINT32

0

Number of centroids. This parameter overrides the general parameter but is overridden by the suggested K value.

proxima.minibatchkmeans.cluster.shard_factor

FLOAT

16.0f

Tuning factor for multi-threaded concurrency.

proxima.minibatchkmeans.cluster.epsilon

DOUBLE

FL_EPSILON

Clustering convergence precision.

proxima.minibatchkmeans.cluster.max_iterations

UINT32

20

Maximum number of iterations.

proxima.minibatchkmeans.cluster.purge_empty

BOOL

false

Specifies whether to delete empty centroids.

proxima.minibatchkmeans.cluster.try_count

UINT32

20

Number of tries. The minimum value is 1.

proxima.minibatchkmeans.cluster.batch_count

UINT32

0 (auto)

The number of sampled features for batch training. If this parameter is set to 0, the value is calculated as total number of features / number of tries.

proxima.minibatchkmeans.cluster.seeker_class

STRING

LinearSeeker

The algorithm class for finding centroids.

proxima.minibatchkmeans.cluster.seeker_params

IndexParams

Parameters for the centroid finding algorithm class.

1.4 BikmeansCluster

Parameter name

Type

Default value

Note

proxima.general.cluster.count

UINT32

0

Number of centroids.

proxima.bikmeans.cluster.count

UINT32

0

Number of centroids. This parameter overrides the general parameter but is overridden by the suggested K value.

proxima.bikmeans.cluster.init_count

UINT32

0 (auto)

The number of centroids for initialization in the first stage of clustering. If this parameter is set to 0, the value is calculated as the total number of features divided by four.

proxima.bikmeans.cluster.purge_empty

BOOL

false

Specifies whether to delete empty centroids.

proxima.bikmeans.cluster.first_class

STRING

KmeansCluster

The clustering method for the first stage.

proxima.bikmeans.cluster.second_params

IndexParams

Parameters for the first-stage clustering method.

proxima.bikmeans.cluster.second_class

STRING

KmeansCluster

The clustering method for the second stage.

proxima.bikmeans.cluster.second_params

IndexParams

Parameters for the second-stage clustering method.

1.5 KmeansppCluster

Parameter

Type

Default value

Note:

proxima.general.cluster.count

UINT32

0

Number of centroids.

proxima.kmeanspp.cluster.count

UINT32

0

Number of centroids. This parameter overrides the general parameter but is overridden by the suggested K value.

proxima.kmeanspp.cluster.shard_factor

UINT32

16.0f

Tuning factor for multi-threaded concurrency.

proxima.kmeanspp.cluster.class

STRING

KmeansCluster

The clustering method to call after centroid initialization.

proxima.kmeanspp.cluster.params

IndexParams

Parameters for the clustering method.

1.6 Kmc2Cluster / AFKmc2Cluster

Parameter

Type

Default value

Note:

proxima.general.cluster.count

UINT32

0

Number of centroids.

proxima.kmc2.cluster.count

UINT32

0

Number of centroids. This parameter overrides the general parameter but is overridden by the suggested K value.

proxima.kmc2.cluster.shard_factor

UINT32

2.5f

Tuning factor for multi-threaded concurrency.

proxima.kmc2.cluster.markov_chain_length

UINT32

0u

The length of the Markov chain. If this parameter is set to 0, the value is calculated as number of threads × concurrency factor.

proxima.kmc2.cluster.class

STRING

KmeansCluster

The clustering method to call after centroid initialization.

proxima.kmc2.cluster.params

IndexParams

Parameters for the clustering method.

1.7 KmedoidsCluster

Parameter name

Type

Default value

Description

proxima.general.cluster.count

UINT32

0

Number of centroids.

proxima.kmedoids.cluster.count

UINT32

0

Number of centroids. This parameter overrides the general parameter but is overridden by the suggested K value.

proxima.kmedoids.cluster.shard_factor

FLOAT

16.0f

Tuning factor for multi-threaded concurrency.

proxima.kmedoids.cluster.epsilon

DOUBLE

FL_EPSILON

Clustering convergence precision.

proxima.kmedoids.cluster.max_iterations

UINT32

20

Maximum number of iterations.

proxima.kmedoids.cluster.purge_empty

BOOL

false

Specifies whether to delete empty centroids.

proxima.kmedoids.cluster.bench_ratio

FLOAT

0.1f

Ratio of candidate points.

proxima.kmedoids.cluster.only_means

BOOL

false

Considers only the mean as a candidate point. The algorithm degrades to k-means.

proxima.kmedoids.cluster.without_means

BOOL

false

Does not consider the mean as a candidate point.

proxima.kmedoids.cluster.seeker_class

STRING

LinearSeeker

The algorithm class for finding centroids.

proxima.kmedoids.cluster.seeker_params

IndexParams

Parameters for the centroid finding algorithm class.

IndexParams object

1.8 StratifiedCluster

Parameter

Type

Default value

Note:

proxima.general.cluster.count

UINT32

0

Total number of second-layer centroids.

proxima.stratified.cluster.count

UINT32

0

Total number of second-layer centroids. This parameter overrides the general parameter but is overridden by the suggested K value.

proxima.stratified.cluster.first_class

STARING

KmeansCluster

The clustering method for the first layer.

proxima.stratified.cluster.second_class

STARING

KmeansCluster

The clustering method for the second layer.

proxima.stratified.cluster.first_count

UINT32

0

Number of first-layer centroids.

proxima.stratified.cluster.second_count

UINT32

0

Number of second-layer centroids.

proxima.stratified.cluster.first_params

IndexParams

Parameters for the first-layer clustering method.

proxima.stratified.cluster.second_params

IndexParams

Parameters for the second-layer clustering method.

proxima.stratified.cluster.auto_tuning

BOOL

false

2. Clustering estimation

2.1 GapstatsClusterEstimator