In [1]:

    

import pandas as pd
import numpy as np
import scipy.stats as scs
from skimage.io import imread
import util
import matplotlib.pyplot as plt
from mpl_toolkits.axes_grid1 import ImageGrid
import os
%matplotlib inline


    

In [15]:

    

train_data = []
test_data = []
test_names = []
train_names = []
for root, files in util.get_person_images('./lfw_funneled/', 'jpg', 70):
    name, data, amount = util.get_dataset(root, files, 8)
    # save name vector
    test_names.append(name)
    train_names.extend([name]*(amount-1))
    train_data.append(data[:-1]) # without last
    test_data.append(data[-1:]) # only last


    

In [23]:

    

train_design_matrix = np.asmatrix(np.concatenate(train_data))
test_design_matrix = np.asmatrix(np.concatenate(test_data))


    

In [34]:

    

test_names, train_design_matrix.shape, test_design_matrix.shape


    

    
Out[34]:





(['Colin_Powell',
  'George_W_Bush',
  'Hugo_Chavez',
  'Ariel_Sharon',
  'Tony_Blair',
  'Gerhard_Schroeder',
  'Donald_Rumsfeld'],
 (1281, 1024),
 (7, 1024))

In [33]:

    

# Plot the test data
fig = plt.figure()
for i, row in enumerate(test_design_matrix):
    ax = plt.subplot(3,3,i+1)
    ax.set_title("{}".format(test_names[i]))
    plt.axis('off')
    plt.tight_layout()
    dim = np.sqrt(row.shape[1])
    plt.imshow(np.reshape(row, (dim, dim)), cmap=plt.gray(), )


    

In [36]:

    

def align(mat):
    mean = mat.mean()


    

In [45]:

    

train_mean = np.mean(train_design_matrix, axis=1)
test_mean = np.mean(test_design_matrix, axis=1)
train_aligned = train_design_matrix - train_mean
test_aligned = test_design_matrix - test_mean


    

In [58]:

    

u, d, v = np.linalg.svd(train_aligned, full_matrices=False)
u.shape, d.shape, v.shape


    

    
Out[58]:





((1281, 1024), (1024,), (1024, 1024))

In [61]:

    

# Plot the first 12 eigenfaces
fig = plt.figure()
for i, eigenface in enumerate(v[:12], start=1):
    plt.subplot(3, 4, i)
    plt.tight_layout()
    dim = np.sqrt(eigenface.shape[1])
    plt.imshow(np.reshape(eigenface, (dim, dim)), cmap=plt.gray())


    

In [62]:

    

eigenvalues = d**2
eigenvalues
# Plot first 150 Eigenvalues
plt.plot(eigenvalues[:150])
plt.show()


    

In [63]:

    

features = 7
train_projected = np.dot(v[:features], train_aligned.T).T
test_projected = np.dot(v[:features], test_aligned.T).T


    

In [67]:

    

print("Shape Projiziert: train {}, test {}".format(np.shape(train_projected), np.shape(test_projected)))


    

    




Shape Projiziert: train (1281, 7), test (7, 7)

In [68]:

    

print("Shape V:", v[:7].shape, " Shape test zentriert:", test_aligned.shape, " Shape Training zentriert:", train_aligned.shape)


    

    




Shape V: (7, 1024)  Shape test zentriert: (7, 1024)  Shape Training zentriert: (1281, 1024)

In [104]:

    




    

    
Out[104]:





((7, 3969), (1281, 3969))

In [ ]:

    




    

In [104]:

    

result = {'Test': [], 'Klassifikation' : [], 'Index': []}

for index, a in enumerate(test_projected):
    # subtract each projected test data from all projected learned data
    subtr = train_projected - a
    # calculate the length of each subtracted data
    distances = np.linalg.norm(subtr, axis=1)
    # take the data with minimal length
    id = distances.argmin()
    name_classification = train_names[id]
    result['Test'].append(names[index])
    result['Klassifikation'].append(name_classification)
    result['Index'].append(id)


    

In [105]:

    

pd.DataFrame(result)


    

    
Out[105]:






  

    

      

      
Index
      
Klassifikation
      
Test
    
  
  

    

      
0
      
1018
      
Tony_Blair
      
Colin_Powell
    
    

      
1
      
654
      
George_W_Bush
      
George_W_Bush
    
    

      
2
      
798
      
Hugo_Chavez
      
Hugo_Chavez
    
    

      
3
      
843
      
Ariel_Sharon
      
Ariel_Sharon
    
    

      
4
      
1090
      
Gerhard_Schroeder
      
Tony_Blair
    
    

      
5
      
662
      
George_W_Bush
      
Gerhard_Schroeder
    
    

      
6
      
182
      
Colin_Powell
      
Donald_Rumsfeld
    
  

In [100]:

    




    

    
Out[100]:





1.0179602570000652

In [ ]: