-
Notifications
You must be signed in to change notification settings - Fork 0
/
svd.js
52 lines (47 loc) · 1.84 KB
/
svd.js
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
// Visual SVD
// Scott Clayton
var svd = {
learningRate: 0.02,
totalIterations: 0,
rmse: 0,
factorize: function (matrix, featWide, featTall, predictionMatrix, errorMatrix, learningIterations) {
var regularizationTerm = 0.01;
var features = featWide[0].length;
for (var i = 0; i < learningIterations; i++) {
var squaredError = 0.0;
var count = 0;
for (var row = 0; row < matrix.length; row++) {
for (var col = 0; col < matrix[0].length; col++) {
if (matrix[row][col] != 0) {
var prediction = svd.dotProduct(featWide[row], featTall[col]);
var error = matrix[row][col] - prediction;
predictionMatrix[row][col] = prediction;
errorMatrix[row][col] = error;
squaredError += error * error;
count++;
for (var feat = 0; feat < features; feat++)
{
featWide[row][feat] += svd.learningRate * (error * featTall[col][feat] - regularizationTerm * featWide[row][feat]);
featTall[col][feat] += svd.learningRate * (error * featWide[row][feat] - regularizationTerm * featTall[col][feat]);
}
} else {
// Predict unknown cells
predictionMatrix[row][col] = svd.dotProduct(featWide[row], featTall[col]);
errorMatrix[row][col] = 0;
}
}
}
svd.totalIterations++;
svd.rmse = Math.sqrt(squaredError / count);
console.log("Iteration: " + svd.totalIterations + " Error: " + svd.rmse);
svd.learningRate *= 0.99;
}
},
dotProduct: function (left, right) {
var prod = 0;
for (var i = 0; i < left.length; i++) {
prod += left[i] * right[i];
}
return prod;
}
};