%matplotlib inline

import matplotlib
import numpy as np
import matplotlib.pyplot as plt
from sklearn.cluster import KMeans

# Reads the adjacency matrix from file

A = np.loadtxt('adjacency.txt')
print(f'There are {A.shape[0]} nodes in the graph.')

There are 328 nodes in the graph.

As you can see above, the adjacency matrix is relatively large (328x328): there are 328 persons in
the graph. In order to visualize this adjacency matrix, it is convenient to use the 'imshow'
function. This plots the 328x328 image where the pixel (i,j) is black if and only if A[i,j]=1.

plt.figure(figsize=(8,8))
plt.imshow(A,aspect='equal',cmap='Greys', interpolation='none')

<matplotlib.image.AxesImage at 0x1a6e5b17a30>
(a) Construct in the cell below the degree matrix:

$$ D_{i,i} = {\rm deg}(i) \qquad \text{and} \qquad D_{i,j} = 0 \ \ \text{if} \ i \neq j, $$

the Laplacian matrix:

L=D − A
and the normalized Laplacian matrix:

$$ L_{\rm norm} = D^{-1/2} L D^{-1/2}. $$

# Your answer here

Dmatrix = np.diag(A.sum(axis=1))
print("Degree Matrix, D :\n", Dmatrix)

# Laplacian Matrix
L = Dmatrix - A
print("\nLaplacian Matrix, L :\n", L)

# Normalized Laplacian Matrix

Lnorm = (np.diag(1.0/ np.sqrt(Dmatrix.diagonal()))) @ L @
(np.diag(1.0/ np.sqrt(Dmatrix.diagonal())))
print("\nNormalized Laplacian Matrix :\n", Lnorm)

Degree Matrix, D :
[[ 27. 0. 0. ... 0. 0. 0.]
[ 0. 83. 0. ... 0. 0. 0.]
[ 0. 0. 65. ... 0. 0. 0.]
...
[ 0. 0. 0. ... 66. 0. 0.]
[ 0. 0. 0. ... 0. 58. 0.]
[ 0. 0. 0. ... 0. 0. 327.]]

Laplacian Matrix, L :
[[ 27. 0. 0. ... 0. 0. -1.]
[ 0. 83. 0. ... 0. 0. -1.]
[ 0. 0. 65. ... 0. -1. -1.]
...
[ 0. 0. 0. ... 66. 0. -1.]
[ 0. 0. -1. ... 0. 58. -1.]
[ -1. -1. -1. ... -1. -1. 327.]]

Normalized Laplacian Matrix :

[[ 1. 0. 0. ... 0. 0.
-0.01064251]
[ 0. 1. 0. ... 0. 0.
-0.00606998]
[ 0. 0. 1. ... 0. -0.01628656
-0.00685914]
...
[ 0. 0. 0. ... 1. 0.
-0.00680698]
[ 0. 0. -0.01628656 ... 0. 1.
-0.00726126]
[-0.01064251 -0.00606998 -0.00685914 ... -0.00680698 -0.00726126
1. ]]

(b) Using the command 'linalg.eigh' from numpy, compute the eigenvalues and the eigenvectors
of $L_{\rm norm}$.

# Your answer here

evalues, evectors = np.linalg.eigh(Lnorm)
print("Eigenvalues : \n", evalues)
print("\n Eigenvectors : \n", evectors)

Eigenvalues :
[-4.32777954e-16 8.22971080e-02 2.73920284e-01 2.86756239e-01
4.12340841e-01 4.41921035e-01 6.16054279e-01 6.70249866e-01
7.06406288e-01 7.25463028e-01 7.35075504e-01 7.52268234e-01
7.71169767e-01 7.73015222e-01 7.88819638e-01 7.93830094e-01
8.03472018e-01 8.14921509e-01 8.21827484e-01 8.27206976e-01
8.35085854e-01 8.38589162e-01 8.46760685e-01 8.56016240e-01
8.58888980e-01 8.61157975e-01 8.65395654e-01 8.68032309e-01
8.71228049e-01 8.75612490e-01 8.79472847e-01 8.81071746e-01
8.83567703e-01 8.85609635e-01 8.87491226e-01 8.90039358e-01
8.93125892e-01 8.96071979e-01 8.97872008e-01 9.00108767e-01
9.02215754e-01 9.04086375e-01 9.06058925e-01 9.07454646e-01
9.09056960e-01 9.09610504e-01 9.13082816e-01 9.13504100e-01
9.15522027e-01 9.16340436e-01 9.17403135e-01 9.19270844e-01
9.22130905e-01 9.23161207e-01 9.23511054e-01 9.24832666e-01
9.28088113e-01 9.29143852e-01 9.30923431e-01 9.32402646e-01
9.33870511e-01 9.35555401e-01 9.36692116e-01 9.38080974e-01
9.39859181e-01 9.41188489e-01 9.42451551e-01 9.44524080e-01
9.45495347e-01 9.46814646e-01 9.47076227e-01 9.48401324e-01
9.48702506e-01 9.50156125e-01 9.50672500e-01 9.51089565e-01
9.52229499e-01 9.54700979e-01 9.55973606e-01 9.57012037e-01
9.59007875e-01 9.59340528e-01 9.60095483e-01 9.61813213e-01
9.62496471e-01 9.63666669e-01 9.63774402e-01 9.64619656e-01
9.65020720e-01 9.65381367e-01 9.66932735e-01 9.67219909e-01
9.69151813e-01 9.69662636e-01 9.70255098e-01 9.71038681e-01
9.71844822e-01 9.72518729e-01 9.73098475e-01 9.74397971e-01
9.74765109e-01 9.76393231e-01 9.77111815e-01 9.77513934e-01
9.78096613e-01 9.79256225e-01 9.79893301e-01 9.80091360e-01
9.81342758e-01 9.81868729e-01 9.82715996e-01 9.83755570e-01
9.84490267e-01 9.85156265e-01 9.85601087e-01 9.86227000e-01
9.87272162e-01 9.87546526e-01 9.88666227e-01 9.88890536e-01
9.89439372e-01 9.89939160e-01 9.90497496e-01 9.91449441e-01
9.92825820e-01 9.93252781e-01 9.94124653e-01 9.94843424e-01
9.95395267e-01 9.96278082e-01 9.97150885e-01 9.98352354e-01
9.98574988e-01 9.98971618e-01 1.00000000e+00 1.00000000e+00
1.00000000e+00 1.00000000e+00 1.00000000e+00 1.00000000e+00
1.00000000e+00 1.00000000e+00 1.00000000e+00 1.00000000e+00
1.00000000e+00 1.00029329e+00 1.00060897e+00 1.00123040e+00
1.00217855e+00 1.00264326e+00 1.00318027e+00 1.00360296e+00
1.00394648e+00 1.00452451e+00 1.00493913e+00 1.00585384e+00
1.00682744e+00 1.00754310e+00 1.00846734e+00 1.00910240e+00
1.00984873e+00 1.01066596e+00 1.01081159e+00 1.01131858e+00
1.01173508e+00 1.01252273e+00 1.01264331e+00 1.01317383e+00
1.01406340e+00 1.01495706e+00 1.01529273e+00 1.01598070e+00
1.01700814e+00 1.01735886e+00 1.01789071e+00 1.01846635e+00
1.01865757e+00 1.01941545e+00 1.02005446e+00 1.02114281e+00
1.02136189e+00 1.02232889e+00 1.02256850e+00 1.02303408e+00
1.02348028e+00 1.02382565e+00 1.02407066e+00 1.02531368e+00
1.02605812e+00 1.02650922e+00 1.02730736e+00 1.02792683e+00
1.02874064e+00 1.02939545e+00 1.02974518e+00 1.03062856e+00
1.03145498e+00 1.03241489e+00 1.03278909e+00 1.03328519e+00
1.03474330e+00 1.03574221e+00 1.03671091e+00 1.03693224e+00
1.03748780e+00 1.03774168e+00 1.03846617e+00 1.03882506e+00
1.03914011e+00 1.04164351e+00 1.04213483e+00 1.04249023e+00
1.04303488e+00 1.04390826e+00 1.04485040e+00 1.04495977e+00
1.04512557e+00 1.04659637e+00 1.04815093e+00 1.04847570e+00
1.04879178e+00 1.05002517e+00 1.05073620e+00 1.05111870e+00
1.05182571e+00 1.05278393e+00 1.05335259e+00 1.05422111e+00
1.05520493e+00 1.05628715e+00 1.05670284e+00 1.05759106e+00
1.05874514e+00 1.05907374e+00 1.05967862e+00 1.06110234e+00
1.06149099e+00 1.06207963e+00 1.06310380e+00 1.06331023e+00
1.06488980e+00 1.06609391e+00 1.06761350e+00 1.06873409e+00
1.06901424e+00 1.06983455e+00 1.07030620e+00 1.07189652e+00
1.07242659e+00 1.07412998e+00 1.07453284e+00 1.07514340e+00
1.07554058e+00 1.07696311e+00 1.07767367e+00 1.07821995e+00
1.08019425e+00 1.08054105e+00 1.08161381e+00 1.08312387e+00
1.08375073e+00 1.08409038e+00 1.08556125e+00 1.08664060e+00
1.08761192e+00 1.08912356e+00 1.08937828e+00 1.09082270e+00
1.09233753e+00 1.09283179e+00 1.09383110e+00 1.09502089e+00
1.09630653e+00 1.09827775e+00 1.09876234e+00 1.09958891e+00
1.10096741e+00 1.10140510e+00 1.10351237e+00 1.10384381e+00
1.10602352e+00 1.10798429e+00 1.10845139e+00 1.10954290e+00
1.11080279e+00 1.11159317e+00 1.11306899e+00 1.11441063e+00
1.11561448e+00 1.11638074e+00 1.11844457e+00 1.11906802e+00
1.12265873e+00 1.12388268e+00 1.12517735e+00 1.12569225e+00
1.12793035e+00 1.13018630e+00 1.13179435e+00 1.13239995e+00
1.13358140e+00 1.13734581e+00 1.13883209e+00 1.14147333e+00
1.14285714e+00 1.14285714e+00 1.14387864e+00 1.14516077e+00
1.14874599e+00 1.15097199e+00 1.15137460e+00 1.15407050e+00
1.16007708e+00 1.16168594e+00 1.16181259e+00 1.16613952e+00
1.17101907e+00 1.17472079e+00 1.17527485e+00 1.19182376e+00
1.19796968e+00 1.20891304e+00 1.23304226e+00 1.27349456e+00
1.28324671e+00 1.39234807e+00 1.41442916e+00 1.57937217e+00]

Eigenvectors :
[[ 3.73659565e-02 -8.21492875e-02 9.47994044e-04 ... -5.44469096e-03
5.32314013e-03 -1.17415024e-04]
[ 6.55138721e-02 3.34978246e-02 -1.72780496e-02 ... 3.81467899e-05
1.39720393e-03 1.95839484e-03]
[ 5.79763540e-02 3.25034504e-02 7.28978656e-02 ... -1.37581831e-04
9.68810202e-04 -1.28229038e-04]
...
[ 5.84206238e-02 2.87958439e-02 -8.34034886e-02 ... -6.56493088e-04
-8.72425941e-05 -5.93146577e-05]
[ 5.47656465e-02 3.06424242e-02 7.66912615e-02 ... -2.46122914e-04
1.28299897e-03 -1.21988865e-04]
 [ 1.30037346e-01 -2.74479658e-02 8.36567360e-04 ... 2.21492906e-02
-1.04092103e-01 7.72424456e-03]]

(c) We would like to cluster the nodes (i.e. the users) in 3 groups. Using the eigenvectors of $L_{\
rm norm}$, assign to each node a point in R2, exactly as explained in last lecture (also in
'Algorithm 1' of the notes) where you replace L by $L_{\rm norm}$. Plot these points using the
'scatter' function of matplotlib.

# Your answer here

points=evectors[:,1:3]
kmeans= KMeans(n_clusters=3)
label=kmeans.fit_predict(points)

plt.figure(figsize=(8,8))
plt.scatter(points[:,0],points[:,1],c=label)
plt.title("Clustering Nodes in 3 Groups")
plt.xlabel("first eigenvector")
plt.ylabel("second eigenvector")
plt.show()
(d) Using the K-means algorithm (use the built-in function from scikit-learn), cluster the
embeddings in R2 of the nodes in 3 groups.

# Replace ??? by the matrix of the points computed in (c)

# Each row corresponds to a data point
kmeans = KMeans(n_clusters=3, random_state=0).fit(points)
labels = kmeans.labels_
# labels contains the membership of each node 0,1 or 2

# This colors each point of R^2 according to its label

# replace "x/y coordinates" by the coordinates you computed in (c)
plt.scatter(points[:, 0], points[:, 1], alpha=0.7, marker='+',
c=labels)
<matplotlib.collections.PathCollection at 0x1a6e6d3c280>

(e) Re-order the adjacency matrix according to the clusters computed in the previous question.
That is, reorder the columns and rows of A to obtain a new adjacency matrix (that represents of
course the same graph) such that the n1 nodes of the first cluster correspond to the first n1
rows/columns, the n2 nodes of the second cluster correspond to the next n2 rows/columns, and
the n3 nodes of the third cluster correspond to the last n3 rows/columns. Plot the reordered
adjacency matrix using 'imshow'.

## Your answer here

sortedindices = np.argsort(label)

#reordering the columns and rows to obtain new adjacency matrix

A_new= A[sortedindices, :][:, sortedindices]

plt.figure(figsize=(8,8))
plt.imshow(A_new,cmap='Greys')
plt.title("New Reordered Adjacency Matrix")
plt.show()