In [3]:

    

import numpy as np
import pylab as plt
from scipy.spatial.distance import pdist,squareform
plt.rcParams['font.family']='Serif'
%matplotlib inline

from sklearn import datasets
data = datasets.load_digits()
n = 1000
X = data.data[:n]
Y = data.target[:n]


    

In [4]:

    

from sklearn.manifold import MDS
mds = MDS(dissimilarity='precomputed')
d = squareform(pdist(X))
X_ = mds.fit_transform(d)


    

In [7]:

    

plt.scatter(X_[:,0],X_[:,1],c=Y,linewidths=0)


    

    
Out[7]:





<matplotlib.collections.PathCollection at 0x10455d908>






    

In [40]:

    

mds = MDS()
X_ = mds.fit_transform(X)


    

In [41]:

    

plt.scatter(X_[:,0],X_[:,1],c=Y,linewidths=0)


    

    
Out[41]:





<matplotlib.collections.PathCollection at 0x11e5b8f98>






    

In [42]:

    

# rearrange indices by class label
ind = sorted(np.arange(len(X)),key=lambda i:Y[i])


    

In [43]:

    

pd = pdist(X[ind])
d_ = squareform(sorted(np.arange(len(pd)),key=lambda i:-pd[i]))


    

In [44]:

    

plt.imshow(d_,cmap='Blues')


    

    
Out[44]:





<matplotlib.image.AxesImage at 0x111ac3d68>






    

In [37]:

    

d_sorted = d_[ind]
d_sorted = d_sorted[:,ind]


    

In [38]:

    

plt.imshow(d_sorted[ind],cmap='Blues')


    

    
Out[38]:





<matplotlib.image.AxesImage at 0x11e8a9be0>






    

In [20]:

    

mds = MDS(dissimilarity='precomputed')
X_ = mds.fit_transform(d_)


    

In [21]:

    

plt.scatter(X_[:,0],X_[:,1],c=Y,linewidths=0)


    

    
Out[21]:





<matplotlib.collections.PathCollection at 0x108755160>






    

In [ ]: