it('should return correct weights', function () {
var regFact = 0.1;
var forgetFact = 0.90;
var n = 1000;
var dim = 3;
var linreg = new analytics.RecLinReg({ dim: dim, regFact: regFact, forgetFact: forgetFact });
var X = new la.Matrix({ rows: n, cols: dim });
var yArr = [];
for (var rowN = 0; rowN < n; rowN++) {
var rowArr = [1];
for (var colN = 1; colN < dim; colN++) {
rowArr.push(Math.random());
}
var xi = new la.Vector(rowArr);
var yi = Math.random();
linreg.partialFit(xi, yi);
X.setRow(rowN, xi);
yArr.push(yi);
}
var y = new la.Vector(yArr);
var wgts = linreg.weights;
var expectedWgts = linregWgtsFromMatrixForgetFact(X, y, regFact, forgetFact);
assert.eqtol(wgts[0], expectedWgts[0], tol);
assert.eqtol(wgts[1], expectedWgts[1], tol);
assert.eqtol(wgts[2], expectedWgts[2], tol);
})
it("should achive perfect reconstruction", function () {
var A = new la.Matrix([[10, 20, 110, 20, 10], [3, 4, 13, 63, 1]]);
var pca = new analytics.PCA({ k: 2, iter: 1000 });
pca.fit(A);
var pA = pca.transform(A);
var rA = pca.inverseTransform(pA);
assert(A.minus(rA).frob() < 1e-6);
});
it("should make test number 52", function () {
// import la module
var la = require('qminer').la;
// create a new matrix
var mat = new la.Matrix([[1, 2], [3, 4]]);
// create a JavaScript array out of matrix
var arr = mat.toArray(); // returns an array [[1, 2], [3, 4]]
});
it("should make test number 20", function () {
// import la module
var la = require('qminer').la;
// create the matrix
var mat = new la.Matrix([[1, 2], [3, 1], [-4, 5]]);
// get the second column of the matrix
var col = mat.getCol(1);
});
it("should make test number 19", function () {
// import la module
var la = require('qminer').la;
// create the matrix
var mat = new la.Matrix([[1, 2], [3, 1], [-4, 5]]);
// get the row id of the maximum value of the second column
var maxRow = mat.colMaxIdx(1); // returns the value 2
});
it("should make test number 15", function () {
// import la module
var la = require('qminer').la;
// create the matrix
var mat = new la.Matrix([[1, 2], [3, 4]]);
// get the frobenious norm of the matrix
var frob = mat.frob(); // returns the value Math.sqrt(30)
});
it("should make test number 24", function () {
// import la module
var la = require('qminer').la;
// create a new matrix
var mat = new la.Matrix([[1, -1, 0], [15, 8, 3], [0, 1, 0]]);
// call diag function
var vec = mat.diag(); // returns a vector [1, 8, 0]
});
it("should make test number 3", function () {
// import la module
var la = require('qminer').la;
// create a new matrix
var mat = new la.Matrix([[2, 3], [-2, -2], [-3, 1]]);
// get the value at the index (2, 1)
var value = mat.at(2, 1); // returns the value 1
});
it("should make test number 13", function () {
// import la module
var la = require('qminer').la;
// create a new matrix
var mat = new la.Matrix([[1, 2], [3, 5]]);
// get matrix as string
var text = mat.toString(); // returns `1 2 \n3 5 \n\n`
});
it("should make test number 14", function () {
// import la module
var la = require('qminer').la;
// create the matrix
var mat = new la.Matrix([[1, 2], [0, 3], [-4, 0]]);
// transform the matrix into the sparse form
var spMat = mat.sparse();
});
it("should make test number 22", function () {
// import la module
var la = require('qminer').la;
// create the matrix
var mat = new la.Matrix([[1, 2], [3, 1], [-4, 5]]);
// get the first row of the matrix
var row = mat.getRow(1);
});
it("should make test number 12", function () {
// import la module
var la = require('qminer').la;
// create a new matrix
var mat = new la.Matrix([[3, 4], [4, 15/2]]);
// get the row norms of the matrix
var rowNorms = mat.colNorms(); // returns the vector [5, 17/2]
});
it("should make test number 10", function () {
// import la module
var la = require('qminer').la;
// create a new matrix
var M = new la.Matrix([[1, 2], [-1, -5]]);
// create a new vector
var b = new la.Vector([-1, -6]);
// solve the linear system M*x = b
var x = M.solve(b); // returns vector [1, -1]
});
it("should make test number 6", function () {
// import la module
var la = require('qminer').la;
// create a new matrix
var mat = new la.Matrix([[1, 2], [-1, 5]]);
// create a new vector
var vec = new la.Vector([1, -1]);
//multiply mat and vec
var vec2 = mat.multiplyT(vec); // returns vector [2, 7]
});
it("should make test number 9", function () {
// import la module
var la = require('qminer').la;
// create a matrix
var mat = new la.Matrix([[2, -5], [3, 1]]);
// transpose the matrix
// the return matrix is
// 2 3
// -5 1
var trans = mat.transpose();
});
it("should make test number 49", function () {
// import la module
var la = require('qminer').la;
// create a new matrix
var mat = new la.Matrix([[1, 2], [3, 4]]);
// print the matrix
// each row represents a row in the matrix. For this example:
// 1 2
// 3 4
mat.print();
});
it("should make test number 15", function () {
// import la module
var la = require('qminer').la;
// create a new matrix
var mat = new la.Matrix([[3, 4], [4, 15/2]]);
// normalize the columns of the matrix
// the matrix is going to be of the form:
// 3/5 8/17
// 4/5 15/17
mat.normalizeCols();
});
it("should make test number 25", function () {
// import the modules
var fs = require('qminer').fs;
var la = require('qminer').la;
// create new matrix
var mat = new la.Matrix([[1, 2], [3, 4]]);
// open write stream
var fout = fs.openWrite('mat.dat');
// save matrix and close write stream
mat.save(fout).close();
});
it("should make test number 26", function () {
// import the modules
var fs = require('qminer').fs;
var la = require('qminer').la;
// create an empty matrix
var mat = new la.Matrix();
// open a read stream ('mat.dat' is pre-saved)
var fin = fs.openRead('mat.dat');
// load the matrix
mat.load(fin);
});
it("should make test number 8", function () {
// import la module
var la = require('qminer').la;
// create two matrices
var mat = new la.Matrix([[1, 2], [-1, 5]]);
var mat2 = new la.Matrix([[1, -1], [3, 2]]);
// substract the matrices
// the return matrix is
// 0 3
// -4 3
var diff = mat.minus(mat2);
});
it("should make test number 4", function () {
// import la module
var la = require('qminer').la;
// create a new matrix
var mat = new la.Matrix([[1, 2, 3], [4, 5, 6], [7, 8, 9]]);
var arg = new la.Matrix([[10, 11], [12, 13]]);
mat.put(0, 1, arg);
// updates the matrix to
// 1 10 11
// 4 12 13
// 7 8 9
});
it("should make test number 7", function () {
// import la module
var la = require('qminer').la;
// create two matrices
var mat = new la.Matrix([[1, 2], [-1, 5]]);
var mat2 = new la.Matrix([[1, -1], [3, 2]]);
// add the matrices
// the return matrix is
// 2 1
// 2 7
var sum = mat.plus(mat2);
});
it("should make test number 23", function () {
// import la module
var la = require('qminer').la;
// create a matrix
var mat = new la.Matrix([[1, -3, 2], [9, 2, -4], [-2, 3, 3]]);
// create a vector
var vec = new la.Vector([-3, 2, 2]);
// set the first row of the matrix with the vector
// the changed matrix is now
// -3 2 2
// 9 2 -4
// -2 3 3
mat.setRow(0, vec);
});
it('should find a soft margin and be biased towards positive examples', function () {
X = [[-10, 1],
[-3, 0],
[-20, -1],
[20, 1],
[3, 0],
[10, -1]];
var y = [1, -1, 1, -1, 1, -1];
var matrix = new la.Matrix(X);
matrix = matrix.transpose();
var vec = new la.Vector(y);
// unbalance: positive examples are 1000 times more important
var SVC = new analytics.SVC({ algorithm: "LIBSVM", c: 1e-3, j: 1e+3 });
SVC.fit(matrix, vec);
assert.eqtol(SVC.predict(matrix).minus(new la.Vector([1, 1, 1, 1, 1, 1])).norm(), 0, 1e-6);
});
it('should find a soft margin', function () {
X = [[-10, 1],
[-3, 0],
[-20, -1],
[20, 1],
[3, 0],
[10, -1]];
var y = [1, -1, 1, -1, 1, -1];
var matrix = new la.Matrix(X);
matrix = matrix.transpose();
var vec = new la.Vector(y);
var SVC = new analytics.SVC({ algorithm: "LIBSVM", c: 1e-3 });
SVC.fit(matrix, vec);
assert.eqtol(SVC.predict(matrix).minus(new la.Vector([1, 1, 1, -1, -1, -1])).norm(), 0, 1e-6);
});
it('should find a fit with RBF kernel', function () {
X = [[1, 0],
[0, 1],
[-1, 0],
[0, -1],
[2, 0],
[0, 2],
[-2, 0],
[0, -2]];
var y = [1, 1, 1, 1, -1, -1, -1, -1];
var matrix = new la.Matrix(X);
matrix = matrix.transpose();
var vec = new la.Vector(y);
var SVC = new analytics.SVC({ algorithm: "LIBSVM", kernel: "RBF", p:10e-6 });
SVC.fit(matrix, vec);
assert.eqtol(SVC.predict(matrix).minus(new la.Vector([1, 1, 1, 1, -1, -1, -1, -1])).norm(), 0, 1e-6);
});
it('should find a fit with polynomial kernel', function () {
X = [[-3],
[-2],
[-1],
[0],
[1],
[2],
[3]];
var y = [9, 4, 1, 0, 4, 1, 9];
var matrix = new la.Matrix(X);
matrix = matrix.transpose();
var vec = new la.Vector(y);
// unbalance: positive examples are 1000 times more important
var SVR = new analytics.SVR({ algorithm: "LIBSVM", kernel: "POLY", degree: 2, p: 10e-3 });
SVR.fit(matrix, vec);
assert.eqtol(SVR.predict(matrix).minus(new la.Vector([9, 4, 1, 0, 1, 4, 9])).norm(), 0, 1e-1);
});
it('should find a fit with RBF kernel', function () {
X = [[1, 0],
[0, 1],
[-1, 0],
[0, -1],
[2, 0],
[0, 2],
[-2, 0],
[0, -2]];
var y = [1, 1, 1, 1, 2, 2, 2, 2];
var matrix = new la.Matrix(X);
matrix = matrix.transpose();
var vec = new la.Vector(y);
// unbalance: positive examples are 1000 times more important
var SVR = new analytics.SVR({ algorithm: "LIBSVM", kernel: "RBF", p: 10e-3 });
SVR.fit(matrix, vec);
assert.eqtol(SVR.predict(matrix).minus(new la.Vector([1, 1, 1, 1, 2, 2, 2, 2])).norm(), 0, 1e-1);
});
it('should fit a model on iris dataset (dense), class=setosa, features:sepal length, sepal width', function () {
var X = require('./irisX.json');
var y = require('./irisY.json');
var matrix = new la.Matrix(X);
matrix = matrix.transpose();
var vec = new la.Vector(y);
var SVC = new analytics.SVC({ algorithm: "LIBSVM", c: 10000 });
SVC.fit(matrix, vec);
// based on matlab
//load fisheriris
//X = [meas(:,1), meas(:,2)];
//Y = nominal(ismember(species,'setosa'));
//y = 2*(double(Y)-1.5);
//s=fitcsvm(X,y,'Standardize',false, 'BoxConstraint',10000);
// weights = s.SupportVectors'* (s.Alpha.*(y(s.IsSupportVector)))
// bias = s.Bias
assert.eqtol(SVC.weights.minus(new la.Vector([-8.5680, 7.1408])).norm(), 0, 1e-3);
assert.eqtol(Math.abs(SVC.bias - 23.1314), 0, 1e-3);
});
it('should find a fit with polynomial kernel', function () {
X = [[-3, 5],
[-1, 0.5],
[0, -0.5],
[1, 0.5],
[3, 5],
[-2, 5],
[-0.5, 1],
[0, 0.5],
[0.5, 1],
[2, 5]];
var y = [1, 1, 1, 1, 1, -1, -1, -1, -1, -1];
var matrix = new la.Matrix(X);
matrix = matrix.transpose();
var vec = new la.Vector(y);
var SVC = new analytics.SVC({ algorithm: "LIBSVM", kernel: "POLY", degree: 2, p:10e-6, eps:10e-6 });
SVC.fit(matrix, vec);
assert.eqtol(SVC.predict(matrix).minus(new la.Vector([1, 1, 1, 1, 1, -1, -1, -1, -1, -1])).norm(), 0, 1e-6);
});