class static
analytics.DpMeans

Source: analyticsdoc.js:1942

DpMeans Clustering is an iterative, data-partitioning algorithm that assigns observations into clusters with the nearest centroid according to some metric. The number of clusters is not known in advance, but can be upper and lower bounded. Rather the parameter is the maximum radius of a cluster lambda.

Properties

centroids
idxv
medoids
relMeanCentroidDist

Methods

explain(X)
fit(X)
getModel()
getParams()
permuteCentroids(mapping)
predict(A)
save(fout)
setParams(params)
transform(A)

new DpMeans([arg])

DpMeans Clustering

Example 

// import analytics and la modules
var analytics = require('qminer').analytics;
var la = require('qminer').la;
// create a KMeans object
var dpmeans = new analytics.DpMeans();
// create the matrix to be fitted
var X = new la.Matrix([[1, -2, -1], [1, 1, -3]]);
// create the model
dpmeans.fit(X);
// predict where the columns of the matrix will be assigned
var Y = new la.Matrix([[1, 1, 0], [-2, 3, 1]]);
var prediction = dpmeans.predict(Y);

Parameter

Name Type Optional Description

arg

(module:analytics~DpMeansParam or module:fs.FIn)

Yes

Construction arguments. There are two ways of constructing:
1. Using the module:analytics~DpMeansParam object,
2. using the file input stream module:fs.FIn.

Properties

centroids

The centroids created with the fit method. Type module:la.Matrix.

Example 

// import the modules
var analytics = require('qminer').analytics;
var la = require('qminer').la;
// create a new DpMeans object
var DpMeans = new analytics.DpMeans({ iter: 1000, lambda: 3 });
// create a matrix to be fitted
var X = new la.Matrix([[1, -2, -1], [1, 1, -3]]);
// create the model with the matrix X
DpMeans.fit(X);
// get the centroids
var centroids = DpMeans.centroids;
// print the first centroid
console.log(centroids.getCol(0));

idxv

The integer vector containing the cluster ids of the training set created with the fit method. Type module:la.IntVector.

Example 

// import the modules
var analytics = require('qminer').analytics;
// create a new DpMeans object
var DpMeans = new analytics.DpMeans({ iter: 1000, lambda: 3 });
// get the idxv
var idxv = DpMeans.idxv;

medoids

The medoids created with the fit method. Type module:la.IntVector.

Example 

// import the modules
var analytics = require('qminer').analytics;
// create a new DpMeans object
var DpMeans = new analytics.DpMeans({ iter: 1000, lambda: 3 });
// get the centroids
var medoids = DpMeans.medoids;

relMeanCentroidDist

Returns the normalized weighted distance between the vectors and their centroids using the following formula: d = \frac{sum_i p_i*sum_j d(x_j,c_i) / n_i}{sum_{k=1}^n d(x_k, mu) / n} = \frac{sum_{i,j} d(c_i,x_j)}{sum_k d(x_k, mu)}

Returns

number relMeanDist

Methods

explain(X) → Array of module:analytics~DpMeansExplain

Returns the IDs of the nearest medoid for each example.

Example 

// import analytics module
var analytics = require('qminer').analytics;
// import linear algebra module
var la = require('qminer').la;
// create a new DpMeans object
var DpMeans = new analytics.DpMeans({ iter: 1000, k: 3 });
// create a matrix to be fitted
var X = new la.Matrix([[1, -2, -1], [1, 1, -3]]);
// create the model with the matrix X using the column IDs [0,1,2]
DpMeans.fit(X, [1234,1142,2355]);
// create the matrix of the prediction vectors
var test = new la.Matrix([[2, -1, 1], [1, 0, -3]]);
// predict/explain - return the closest medoids
var explanation = DpMeans.explain(test);

Parameter

Name Type Optional Description

X

(module:la.Matrix or module:la.SparseMatrix)

 

Matrix whose columns correspond to examples.
Returns

Array of module:analytics~DpMeansExplain Array containing the DpMeans explanantions.

fit(X) → module:analytics.DpMeans

Calculates the centroids.

Examples

Asynchronous function 

// import analytics module
var analytics = require('qminer').analytics;
var la = require('qminer').la;
// create a new DpMeans object
var DpMeans = new analytics.DpMeans({ iter: 1000, lambda: 3 });
// create a matrix to be fitted
var X = new la.Matrix([[1, -2, -1], [1, 1, -3]]);
// create the model with the matrix X
DpMeans.fitAsync(X, function (err) {
    if (err) console.log(err);
    // successful calculation
});

Synchronous function 

var analytics = require('qminer').analytics;
var la = require('qminer').la;
// create a new DpMeans object
var DpMeans = new analytics.DpMeans({ iter: 1000, lambda: 3 });
// create a matrix to be fitted
var X = new la.Matrix([[1, -2, -1], [1, 1, -3]]);
// create the model with the matrix X
DpMeans.fit(X);

Parameter

Name Type Optional Description

X

(module:la.Matrix or module:la.SparseMatrix)

 

Matrix whose columns correspond to examples.
Returns

module:analytics.DpMeans Self. The model has been updated.

getModel() → Object

Returns the model.

Example 

// import modules
var analytics = require('qminer').analytics;
var la = require('qminer').la;
// create the KMeans object
var dpmeans = new analytics.DpMeans({ iter: 1000 });
// create a matrix to be fitted
var X = new la.Matrix([[1, -2, -1], [1, 1, -3]]);
// create the model
dpmeans.fit(X);
// get the model
var model = dpmeans.getModel();
Returns

Object The DpMeansModel object containing the properites:
1. DpMeansModel.C - The module:la.Matrix or module:la.SparseMatrix containing the centroids,
2. DpMeansModel.medoids - The module:la.IntVector of cluster medoids of the training data,
3. DpMeansModel.idxv - The module:la.IntVector of cluster IDs of the training data.

getParams() → module:analytics~KMeansParam

Returns the parameters.

Example 

// import analytics module
var analytics = require('qminer').analytics;
// create a new DpMeans object
var DpMeans = new analytics.DpMeans({ iter: 1000, lambda: 5 });
// get the parameters
var json = DpMeans.getParams();
console.log(json.lambda);
Returns

module:analytics~KMeansParam The construction parameters.

permuteCentroids(mapping) → module:analytics.DpMeans

Permutates the clusters, and with it module:analytics.DpMeans#centroids, module:analytics.DpMeans#medoids and module:analytics.DpMeans#idxv.

Example 

// import the modules
var analytics = require('qminer').analytics;
var la = require('qminer').la;
// create a new DpMeans object
var DpMeans = new analytics.DpMeans({ iter: 1000, lambda: 3 });
// create a matrix to be fitted
var X = new la.Matrix([[1, -2, -1], [1, 1, -3]]);
// create the model with the matrix X
DpMeans.fit(X);
if (DpMeans.centroids.cols > 1) {
    // create the mapping vector: swap first two centroids
    var Mapping = new la.IntVector([1, 0, 2, 3, 4, 5].splice(0,DpMeans.centroids.cols));
    console.log(DpMeans.centroids.toString());
    // permutate the clusters.
    DpMeans.permuteCentroids(Mapping);
    console.log(DpMeans.centroids.toString());
}

Parameter

Name Type Optional Description

mapping

module:la.IntVector

 

The mapping, where mapping[4] = 2 means "map cluster 4 into cluster 2".
Returns

module:analytics.DpMeans Self. The clusters have been permuted.

predict(A) → module:la.IntVector

Returns an vector of cluster id assignments.

Example 

// import analytics module
var analytics = require('qminer').analytics;
var la = require('qminer').la;
// create a new DpMeans object
var DpMeans = new analytics.DpMeans({ iter: 1000, lambda: 3 });
// create a matrix to be fitted
var X = new la.Matrix([[1, -2, -1], [1, 1, -3]]);
// create the model with the matrix X
DpMeans.fit(X);
// create the matrix of the prediction vectors
var pred = new la.Matrix([[2, -1, 1], [1, 0, -3]]);
// predict the values
var prediction = DpMeans.predict(pred);

Parameter

Name Type Optional Description

A

(module:la.Matrix or module:la.SparseMatrix)

 

Matrix whose columns correspond to examples.
Returns

module:la.IntVector Vector of cluster assignments.

save(fout) → module:fs.FOut

Saves DpMeans internal state into (binary) file.

Example 

// import the modules
var analytics = require('qminer').analytics;
var la = require('qminer').la;
var fs = require('qminer').fs;
// create a new DpMeans object
var DpMeans = new analytics.DpMeans({ iter: 1000, lambda: 3 });
// create a matrix to be fitted
var X = new la.Matrix([[1, -2, -1], [1, 1, -3]]);
// create the model with the matrix X
DpMeans.fit(X);
// create the file output stream
var fout = new fs.openWrite('DpMeans.bin');
// save the DpMeans instance
DpMeans.save(fout);
fout.close();
// load the DpMeans instance
var fin = fs.openRead('DpMeans.bin');
var KMeans2 = new analytics.DpMeans(fin);

Parameter

Name Type Optional Description

fout

module:fs.FOut

 

The output stream.
Returns

module:fs.FOut The output stream fout.

setParams(params) → module:analytics.DpMeans

Sets the parameters.

Example 

// import analytics module
var analytics = require('qminer').analytics;
// create a new DpMeans object
var DpMeans = new analytics.DpMeans();
// change the parameters of the DpMeans object
DpMeans.setParams({ iter: 1000, lambda: 5 });

Parameter

Name Type Optional Description

params

module:analytics~KMeansParam

 

The construction parameters.
Returns

module:analytics.DpMeans Self. The model parameters have been updated.

transform(A) → module:la.Matrix

Transforms the points to vectors of distances to centroids.

Example 

// import modules
var analytics = require('qminer').analytics;
var la = require('qminer').la;
// create a new DpMeans object
var DpMeans = new analytics.DpMeans({ iter: 1000, lambda: 3 });
// create a matrix to be fitted
var X = new la.Matrix([[1, -2, -1], [1, 1, -3]]);
// create the model with the matrix X
DpMeans.fit(X);
// create the matrix of the transform vectors
var matrix = new la.Matrix([[-2, 0], [0, -3]]);
// get the transform values of matrix
// returns the matrix
//  10    17
//   1    20
//  10     1
DpMeans.transform(matrix);

Parameter

Name Type Optional Description

A

(module:la.Matrix or module:la.SparseMatrix)

 

Matrix whose columns correspond to examples.
Returns

module:la.Matrix Matrix where each column represents the squared distances to the centroid vectors.