Plot multi-dimension cluster to 2D plot python

问题:

I was working on clustering a lot of data, which has two different clusters.

The first type is a 6-dimensional cluster whereas the second type is a 12-dimensional cluster. For now I have decided to use kmeans (as it seems the most intuitive clustering algorithm for the start).

The question is how can I map these clusters on a 2d plot so that I can infer whether kmeans is working or not. I would like to use matplotlib, but any other python package is fine.

Cluster 1 is a cluster made up of these data types (int,float,float,int,float,int)

Cluster 2 is a cluster made up of 12 float types.

Trying to get an output similar to this Any tips will be useful.

回答1:

Well after searching internet and getting lots of weird comment less solutions. I was able to figure out how to do it. Here's the code if you are trying to do something similar. It contains codes from various sources and a lot of them written/edited by me. I hope its easier to understand than others out there.

The function was based on kmeans2 from scipy which returns the centroid_list and label_list. The kmeansdata is the numpy array passed to kmeans2 for clustering and the num_clusters denotes the number of clusters passed to kmeans2.

The code writes back a new png file ensuring it doesn't overwrite something else. Also plots only 50 clusters (If you have 1000's of clusters, then dont try to output all of them)

(It was written for python2.7, should work for other versions too I guess.)

import numpy
import colorsys
import random
import os
from matplotlib.mlab import PCA as mlabPCA
from matplotlib import pyplot as plt


def get_colors(num_colors):
    """
    Function to generate a list of randomly generated colors
    The function first generates 256 different colors and then
    we randomly select the number of colors required from it
    num_colors        -> Number of colors to generate
    colors            -> Consists of 256 different colors
    random_colors     -> Randomly returns required(num_color) colors
    """
    colors = []
    random_colors = []
    # Generate 256 different colors and choose num_clors randomly
    for i in numpy.arange(0., 360., 360. / 256.):
        hue = i / 360.
        lightness = (50 + numpy.random.rand() * 10) / 100.
        saturation = (90 + numpy.random.rand() * 10) / 100.
        colors.append(colorsys.hls_to_rgb(hue, lightness, saturation))

    for i in range(0, num_colors):
        random_colors.append(colors[random.randint(0, len(colors) - 1)])
    return random_colors


def random_centroid_selector(total_clusters , clusters_plotted):
    """
    Function to generate a list of randomly selected
    centroids to plot on the output png
    total_clusters        -> Total number of clusters
    clusters_plotted      -> Number of clusters to plot
    random_list           -> Contains the index of clusters
                             to be plotted
    """
    random_list = []
    for i in range(0 , clusters_plotted):
        random_list.append(random.randint(0, total_clusters - 1))
    return random_list

def plot_cluster(kmeansdata, centroid_list, label_list , num_cluster):
    """
    Function to convert the n-dimensional cluster to 
    2-dimensional cluster and plotting 50 random clusters
    file%d.png    -> file where the output is stored indexed
                     by first available file index
                     e.g. file1.png , file2.png ...
    """
    mlab_pca = mlabPCA(kmeansdata)
    cutoff = mlab_pca.fracs[1]
    users_2d = mlab_pca.project(kmeansdata, minfrac=cutoff)
    centroids_2d = mlab_pca.project(centroid_list, minfrac=cutoff)


    colors = get_colors(num_cluster)
    plt.figure()
    plt.xlim([users_2d[:, 0].min() - 3, users_2d[:, 0].max() + 3])
    plt.ylim([users_2d[:, 1].min() - 3, users_2d[:, 1].max() + 3])

    # Plotting 50 clusters only for now
    random_list = random_centroid_selector(num_cluster , 50)

    # Plotting only the centroids which were randomly_selected
    # Centroids are represented as a large 'o' marker
    for i, position in enumerate(centroids_2d):
        if i in random_list:
            plt.scatter(centroids_2d[i, 0], centroids_2d[i, 1], marker='o', c=colors[i], s=100)


    # Plotting only the points whose centers were plotted
    # Points are represented as a small '+' marker
    for i, position in enumerate(label_list):
        if position in random_list:
            plt.scatter(users_2d[i, 0], users_2d[i, 1] , marker='+' , c=colors[position])

    filename = "name"
    i = 0
    while True:
        if os.path.isfile(filename + str(i) + ".png") == False:
            #new index found write file and return
            plt.savefig(filename + str(i) + ".png")
            break
        else:
            #Changing index to next number
            i = i + 1
    return

回答2:

   plot_cluster(X[:], kmean.cluster_centers_, kmean.labels_, clusters)