Using python to generate clusters of data?

Question

I'm working on a Python function, where I want to model a Gaussian distribution, I'm stuck though.

import numpy.random as rnd
import numpy as np

def genData(co1, co2, M):
  X = rnd.randn(2, 2M + 1)
  t = rnd.randn(1, 2M + 1)
  numpy.concatenate(X, co1)
  numpy.concatenate(X, co2)
  return(X, t)

I'm trying for two clusters of size M, cluster 1 is centered at co1, cluster 2 is centered at co2. X would return the data points I'm going to graph, and t are the target values (1 if cluster 1, 2 if cluster 2) so I can color it by cluster.

In that case, t is size 2M of 1s/2s and X is size 2M * 1, wherein t[i] is 1 if X[i] is in cluster 1 and the same for cluster 2.

I figured the best way to start doing this is generating the array array using numpys random. What I'm confused about is how to get it centered according to the cluster?

---

Would the best way be to generate a cluster sized M, then add co1 to each of the points? How would I make it random though, and make sure t[i] is colored in properly?

I'm using this function to graph the data:

def graphData():
    co1 = (0.5, -0.5)
    co2 = (-0.5, 0.5)
    M = 1000
    X, t = genData(co1, co2, M)
    colors = np.array(['r', 'b'])
    plt.figure()
    plt.scatter(X[:, 0], X[:, 1], color = colors[t], s = 10)

Doomsday

Use numpy.random.multivariate_normal. Give the mean argument as a vector of length 2; that will be the location of the cluster. – Hargett

@WarrenWeckesser Thanks Warren, but how will I make it so X is random and t will tell me which cluster it belongs to? – Doomsday

Answer 1

For your purpose, I would go for sklearn sample generator make_blobs:

from sklearn.datasets import make_blobs

centers = [(-5, -5), (5, 5)]
cluster_std = [0.8, 1]

X, y = make_blobs(n_samples=100, cluster_std=cluster_std, centers=centers, n_features=2, random_state=1)

plt.scatter(X[y == 0, 0], X[y == 0, 1], color="red", s=10, label="Cluster1")
plt.scatter(X[y == 1, 0], X[y == 1, 1], color="blue", s=10, label="Cluster2")

You can generate multi-dimensional clusters with this. X yields data points and y is determining which cluster a corresponding point in X belongs to.

This might be too much for what you try to achieve in this case, but generally, I think it's better to rely on more general and better-tested library codes that can be used in other cases as well.

Samuelson

Works great. The only minor thing is samples_generator is now deprecated. Should use from sklearn.datasets import make_blobs instead. – Jarrell

Answer 2

You can use something like following code:

center1 = (50, 60)
center2 = (80, 20)
distance = 20


x1 = np.random.uniform(center1[0], center1[0] + distance, size=(100,))
y1 = np.random.normal(center1[1], distance, size=(100,)) 

x2 = np.random.uniform(center2[0], center2[0] + distance, size=(100,))
y2 = np.random.normal(center2[1], distance, size=(100,)) 

plt.scatter(x1, y1)
plt.scatter(x2, y2)
plt.show()

Salpingotomy