1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
|
# -*- coding: utf-8 -*-
"""
Created on Fri Sep 11 13:12:03 2015
@author: Camil Staps, s4498062
Use Python 2.*
"""
from __future__ import print_function
import xlrd
import numpy as np
import numpy.linalg as la
import matplotlib.lines as pltlines
import matplotlib.patches as pltpatches
import matplotlib.pyplot as plt
# 1.2.1 a
xls = xlrd.open_workbook(filename='Data/nanonose.xls')
xls = xls.sheet_by_index(0)
fst_col, fst_row = 3, 2
data = np.asmatrix([xls.col_values(i)[fst_row:]
for i in range(fst_col,fst_col + 8)])
# 1.2.1 b
colors = {'Water': '#61d4fa',
'Ethanol': '#ff3333',
'Acetone': '#549900',
'Heptane': '#d9910d',
'Pentanol': '#990096'}
graph_colors = [colors[r] for r in xls.col_values(0)[fst_row:]]
xs = xls.col_values(1)[fst_row:]
fig = plt.figure(figsize=(12,6))
ax = plt.gca()
ax.set_xscale('log')
ax.set_yscale('symlog')
plt.xlim([80,10 ** 5])
plt.ylim([-1, 500])
ax.scatter(xs, data.tolist()[0], s=60, c=graph_colors, alpha=0.4, marker='s')
ax.scatter(xs, data.tolist()[1], s=60, c=graph_colors, alpha=0.4, marker='o')
line_a = pltlines.Line2D([], [], ls=' ', marker='s', label='A', c='w')
line_b = pltlines.Line2D([], [], ls=' ', marker='o', label='B', c='w')
handles = [pltpatches.Patch(label=k, color=v) for k, v in
colors.iteritems()] + [line_a, line_b]
ax.legend(handles=handles, numpoints=1, loc=2)
plt.show()
# 1.2.2 a
# PCA is a method that can be used to reduce dimensionality of a dataset. It
# can be used when some variables are correlated; we then basically rewrite one
# of them as a function of the other. Of course, in general that implies data
# loss.
# 1.2.2 b
# EVD is a way to rewrite a diagonalizable matrix into a canonical form (a
# summation of products of eigenvalues and corresponding eigenvectors). SVD is
# a generalisation which can be applied to any matrix.
#
# In SVD, we write A = U*S*V^T. The Us are eigenvectors of AA^T (which can be
# found using EVD); the Vs are eigenvectors of A^TA.
# 1.2.2 c
# Subtract mean
means = [np.mean(np.array(data.T)[i]) for i in range(0,8)]
means_matrix = np.transpose([means for _ in range(0, len(np.array(data)[0]))])
normalised_data = np.transpose(data - means_matrix)
# Perform SVD
U, S, V = la.svd(normalised_data)
# Compute component variances
squared_sum = sum([S[m_] ** 2 for m_ in range(1,len(S))])
variance = [100 * S[m] ** 2 / squared_sum for m in range(1,len(S))]
# Plot
fig = plt.figure(figsize=(2,4))
ax = plt.gca()
ax.bar(range(0, len(variance)), variance, 1)
plt.show()
print("The first 3 components account for", sum(variance[:3]), "% variance.")
# 1.2.2 d
V = np.transpose(V)
projections = [normalised_data * V[:,i] for i in range(0,8)]
for i in range(0,8): # Isn't this possible in an easier way?
projections[i] = [np.array(e)[0][0] for s in projections[i] for e in s]
# Plot
fig = plt.figure(figsize=(12,6))
ax = plt.gca()
ax.scatter(xs, projections[0],
c='#ff0000', marker='o', label='Projection onto component 1')
ax.scatter(xs, projections[1],
c='#00ff00', marker='o', label='Projection onto component 2')
handles, labels = ax.get_legend_handles_labels()
ax.legend(handles, labels, loc=2)
plt.show()
print("In the graph above we see that rougly 70% of the variance is accounted "
"for by the first two components. If we would plot only the first two "
"dimensions of the data, we would have on average only 25% accounted for.")
# 1.2.2 e
print(V[:,1])
# As you can see, mainly takes into account the last two attributes (G and H).
# Adherence with attribute A, B or C would give a large negative projection.
# Adherence with attribute G or H would give a large positive projection.
|
