K-Means is a classical unsupervised clustering Learning Algorithm. The detail of the theory about K-Means that you can find it in Wikipedia. Now I introduce to implement this algorithm by myself.

If you are interesting in the implementation and change it into a better version, you could find it in my github repository and give me some advices. I will be appreciated.

So consider about if I want to classify the data into three different cluster. How could I make it?

images

Here is the result:

images

With the mean values:

1
2
3
4
Means =
[[3.5       1       6]
 [1.66666   6       6]]

In the implementation, I just choose the euclidean distance equation as my sensor to calculate the distance between samples. You could assign the self.distance with your function which is in your application.

Here, I show you how to classify the sample point in K-Means.

1
2
3
4
5
6
7
8
9
10
11
12
13
def classify(self):
    for i in range(self.SampleNum):
        minDis = +numpy.inf
        label  = None
        for k in range(self.classNum):
            d = self.distance(self._Mat[:, i].tolist(), self.meanVal[:, k].tolist())
            if d < minDis:
                minDis = d
                label  = k
        self.classification[i][0] = label
        self.classification[i][1] = minDis

And, here you will glance at the main procesure of this algorithm.

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
"""
After you initialized this class, just call this
function and K Means Model will be built
"""
def train(self):
    while True:
        if self.stopOrNot():
            return
        self.classify()
        for k in range(self.classNum):
            mean    = None
            counter = 0
            for i in range(self.SampleNum):
                if self.classification[i][0] == k:
                    if mean == None:
                        mean =  numpy.array(self._Mat[:, i])*1.
                    else:
                        mean += self._Mat[:, i]
                    counter += 1.
            mean /= counter
            self.meanVal[:, k] = mean

Hope my work will help you in some day. Thank you.

Yous, EOF

Photo by Annabella

Aha! Look! What a big shark. I’m fighting …

images