torchsight.evaluators.flickr32.flickrlogos.ConfusionMatrix module
Confusion matrix.
Created on 06.09.2010
@author: Stefan Romberg, stefan.romberg@informatik.uni-augsburg.de
Source code
'''
Confusion matrix.
Created on 06.09.2010
@author: Stefan Romberg, stefan.romberg@informatik.uni-augsburg.de
'''
__version__ = "0.1"
__author__ = "Stefan Romberg"
__all__ = ['ConfusionMatrix']
import numpy as np
import matplotlib.pyplot as plt
#import pylab as P
#===============================================================================
#
#===============================================================================
class ListOfListMat(list):
"""A simple matrix implemented as list of lists."""
__slots__ = {}
def __init__(self, rows, cols):
list.__init__(self)
for i in range(0, rows):
self.append( [0]*cols )
class ConfusionMatrix(object):
"""Confusion Matrix."""
def __init__(self):
"""Creates a new empty confusion matrix."""
self.mat = dict()
def add_result(self, actual_class, detected_class, count=1):
"""Adds one classification results to the confusion matrix."""
if actual_class in self.mat:
row = self.mat[actual_class]
if detected_class in row:
row[detected_class] += count
else:
row[detected_class] = count
else:
row = dict()
row[detected_class] = count
self.mat[actual_class] = row
def _keys(self):
"""Returns a list of column/row names in this confusion matrix."""
key_set = set()
for actual_class,detected_classes in self.mat.items():
key_set.add(actual_class)
for detected_class in detected_classes.keys():
key_set.add(detected_class)
key_set = sorted(list(key_set))
# no-logo class should come first
if "no-logo" in key_set:
key_set.remove( "no-logo" )
tmp = ["no-logo"]
tmp.extend(key_set)
key_set = tmp
return list(key_set)
def aslistoflist(self):
"""Returns the confusion matrix as list-of-lists."""
class_names = self._keys()
n = len(class_names)
idx_map = dict()
for i,k in enumerate(class_names):
idx_map[k] = i
m = ListOfListMat(n, n)
for actual_class,detected_classes in self.mat.items():
for detected_class,count in detected_classes.items():
m[ idx_map[actual_class] ][ idx_map[detected_class] ] = count
return m
def plot(self, title=None):
"""Plots the confusion matrix."""
# conf_arr = [[33,2,0,0,0,0,0,0,0,1,3], [3,31,0,0,0,0,0,0,0,0,0], [0,4,41,0,0,0,0,0,0,0,1], [0,1,0,30,0,6,0,0,0,0,1], [0,0,0,0,38,10,0,0,0,0,0], [0,0,0,3,1,39,0,0,0,0,4], [0,2,2,0,4,1,31,0,0,0,2], [0,1,0,0,0,0,0,36,0,2,0], [0,0,0,0,0,0,1,5,37,5,1], [3,0,0,0,0,0,0,0,0,39,0], [0,0,0,0,0,0,0,0,0,0,38] ]
conf_arr = self.aslistoflist()
class_names = self._keys()
class_nums = range(0, len(class_names))
norm_conf = []
for row_i in conf_arr:
row_sum = float(sum(row_i))
if row_sum > 0:
normalized_row1 = [ float(j)/row_sum for j in row_i ]
normalized_row = []
for j in normalized_row1:
if j > 0:
normalized_row.append( max(j, 0.03) )
else:
normalized_row.append( 0 )
else:
normalized_row = [ 0 for j in row_i ]
norm_conf.append(normalized_row)
#plt.clf()
fig = plt.figure(figsize=(10,8))
ax = fig.add_subplot(111)
mat = np.array(norm_conf)
# flip upside-down
mat = np.flipud(mat)
class_names_r = list(reversed(class_names))
conf_arr = reversed(conf_arr)
res = ax.imshow(mat, cmap=plt.cm.get_cmap("jet"), interpolation='nearest')
cb = plt.colorbar(res, shrink=0.74, #pad=0.3,
#extend="both",
spacing="proportional",
orientation='vertical') #panchor=(1.0, 0.3))
dark_color = (0,0,0)
bright_color = (0.8,0.8,0.8)
for i, cas in enumerate(conf_arr):
for j, c in enumerate(cas):
#p = norm_conf[i][j]
p = mat[i,j]
fcolor = dark_color
if p < 0.35:
fcolor = bright_color
if c == 0:
pass
elif c >= 1 and c <= 9:
plt.text(j-.20, i-.25, c, fontsize=6, color=fcolor)
elif c >= 10 and c <= 99:
plt.text(j-.40, i-.25, c, fontsize=6, color=fcolor)
elif c >= 100 and c <= 999:
plt.text(j-.40, i-.25, c, fontsize=4, color=fcolor)
elif c >= 1000 and c <= 9999:
plt.text(j-.40, i-.25, c, fontsize=3, color=fcolor)
else:
raise Exception("Too big number: "+str(c))
if title is not None:
ax.set_title(title)
ax.set_xlabel("classified as")
ax.set_ylabel("actual class")
ax.set_xlim(-0.5, max(class_nums) + 0.5)
ax.set_ylim(-0.5, max(class_nums) + 0.5)
import matplotlib.lines as mpllines
lines = ax.get_xticklines()
for line in lines:
line.set_marker(mpllines.TICKDOWN)
lines = ax.get_yticklines()
for line in lines:
line.set_marker(mpllines.TICKLEFT)
plt.yticks( class_nums, [ label+" " for label in class_names_r ],
rotation=0, horizontalalignment='right', fontsize=8)
plt.xticks( class_nums, [ label+" " for label in class_names ],
rotation=90, fontsize=8)
plt.minorticks_off()
plt.tick_params(direction="out", left=True, bottom=True, top=False, right=False)
#plt.subplots_adjust(top=0.95, bottom=0.12, left=0.28, right=0.77)
# note:
# needed to re-adjust shrink,pad value as matplotlib changed its behaviour
plt.subplots_adjust(top=0.95, bottom=0.12, left=0.28, right=0.90)
return plt
def __len__(self):
"""Returns the size of the confusion matrix."""
return len(self.mat)
if __name__ == '__main__':
#===========================================================================
# TESTS
#===========================================================================
# run all doctests
import doctest
doctest.testmod()
cm = ConfusionMatrix()
cm.add_result("cocacola", "puma")
cm.add_result("cocacola", "puma")
cm.add_result("cocacola", "puma")
cm.add_result("cocacola", "puma")
cm.add_result("cocacola", "cocacola")
cm.add_result("cocacola", "nike")
cm.add_result("puma", "cocacola")
cm.add_result("puma", "puma")
cm.add_result("puma", "puma")
cm.add_result("puma", "puma")
cm.add_result("puma", "puma")
cm.add_result("puma", "puma")
cm.add_result("dhl", "nike")
cm.add_result("dhl", "dhl")
cm.add_result("dhl", "dhl")
cm.add_result("dhl", "dhl")
cm.add_result("adidas", "adidas")
cm.add_result("adidas", "adidas")
cm.add_result("adidas", "adidas")
cm.add_result("adidas", "adidas")
cm.add_result("adidas", "adidas")
cm.add_result("adidas", "adidas")
cm.add_result("adidas", "adidas")
cm.add_result("adidas", "nike")
cm.add_result("adidas", "puma")
plt = cm.plot()
plt.show()
print("All doctests passed.")Classes
class ConfusionMatrix-
Confusion Matrix.
Source code
class ConfusionMatrix(object): """Confusion Matrix.""" def __init__(self): """Creates a new empty confusion matrix.""" self.mat = dict() def add_result(self, actual_class, detected_class, count=1): """Adds one classification results to the confusion matrix.""" if actual_class in self.mat: row = self.mat[actual_class] if detected_class in row: row[detected_class] += count else: row[detected_class] = count else: row = dict() row[detected_class] = count self.mat[actual_class] = row def _keys(self): """Returns a list of column/row names in this confusion matrix.""" key_set = set() for actual_class,detected_classes in self.mat.items(): key_set.add(actual_class) for detected_class in detected_classes.keys(): key_set.add(detected_class) key_set = sorted(list(key_set)) # no-logo class should come first if "no-logo" in key_set: key_set.remove( "no-logo" ) tmp = ["no-logo"] tmp.extend(key_set) key_set = tmp return list(key_set) def aslistoflist(self): """Returns the confusion matrix as list-of-lists.""" class_names = self._keys() n = len(class_names) idx_map = dict() for i,k in enumerate(class_names): idx_map[k] = i m = ListOfListMat(n, n) for actual_class,detected_classes in self.mat.items(): for detected_class,count in detected_classes.items(): m[ idx_map[actual_class] ][ idx_map[detected_class] ] = count return m def plot(self, title=None): """Plots the confusion matrix.""" # conf_arr = [[33,2,0,0,0,0,0,0,0,1,3], [3,31,0,0,0,0,0,0,0,0,0], [0,4,41,0,0,0,0,0,0,0,1], [0,1,0,30,0,6,0,0,0,0,1], [0,0,0,0,38,10,0,0,0,0,0], [0,0,0,3,1,39,0,0,0,0,4], [0,2,2,0,4,1,31,0,0,0,2], [0,1,0,0,0,0,0,36,0,2,0], [0,0,0,0,0,0,1,5,37,5,1], [3,0,0,0,0,0,0,0,0,39,0], [0,0,0,0,0,0,0,0,0,0,38] ] conf_arr = self.aslistoflist() class_names = self._keys() class_nums = range(0, len(class_names)) norm_conf = [] for row_i in conf_arr: row_sum = float(sum(row_i)) if row_sum > 0: normalized_row1 = [ float(j)/row_sum for j in row_i ] normalized_row = [] for j in normalized_row1: if j > 0: normalized_row.append( max(j, 0.03) ) else: normalized_row.append( 0 ) else: normalized_row = [ 0 for j in row_i ] norm_conf.append(normalized_row) #plt.clf() fig = plt.figure(figsize=(10,8)) ax = fig.add_subplot(111) mat = np.array(norm_conf) # flip upside-down mat = np.flipud(mat) class_names_r = list(reversed(class_names)) conf_arr = reversed(conf_arr) res = ax.imshow(mat, cmap=plt.cm.get_cmap("jet"), interpolation='nearest') cb = plt.colorbar(res, shrink=0.74, #pad=0.3, #extend="both", spacing="proportional", orientation='vertical') #panchor=(1.0, 0.3)) dark_color = (0,0,0) bright_color = (0.8,0.8,0.8) for i, cas in enumerate(conf_arr): for j, c in enumerate(cas): #p = norm_conf[i][j] p = mat[i,j] fcolor = dark_color if p < 0.35: fcolor = bright_color if c == 0: pass elif c >= 1 and c <= 9: plt.text(j-.20, i-.25, c, fontsize=6, color=fcolor) elif c >= 10 and c <= 99: plt.text(j-.40, i-.25, c, fontsize=6, color=fcolor) elif c >= 100 and c <= 999: plt.text(j-.40, i-.25, c, fontsize=4, color=fcolor) elif c >= 1000 and c <= 9999: plt.text(j-.40, i-.25, c, fontsize=3, color=fcolor) else: raise Exception("Too big number: "+str(c)) if title is not None: ax.set_title(title) ax.set_xlabel("classified as") ax.set_ylabel("actual class") ax.set_xlim(-0.5, max(class_nums) + 0.5) ax.set_ylim(-0.5, max(class_nums) + 0.5) import matplotlib.lines as mpllines lines = ax.get_xticklines() for line in lines: line.set_marker(mpllines.TICKDOWN) lines = ax.get_yticklines() for line in lines: line.set_marker(mpllines.TICKLEFT) plt.yticks( class_nums, [ label+" " for label in class_names_r ], rotation=0, horizontalalignment='right', fontsize=8) plt.xticks( class_nums, [ label+" " for label in class_names ], rotation=90, fontsize=8) plt.minorticks_off() plt.tick_params(direction="out", left=True, bottom=True, top=False, right=False) #plt.subplots_adjust(top=0.95, bottom=0.12, left=0.28, right=0.77) # note: # needed to re-adjust shrink,pad value as matplotlib changed its behaviour plt.subplots_adjust(top=0.95, bottom=0.12, left=0.28, right=0.90) return plt def __len__(self): """Returns the size of the confusion matrix.""" return len(self.mat)Methods
def __init__(self)-
Creates a new empty confusion matrix.
Source code
def __init__(self): """Creates a new empty confusion matrix.""" self.mat = dict() def add_result(self, actual_class, detected_class, count=1)-
Adds one classification results to the confusion matrix.
Source code
def add_result(self, actual_class, detected_class, count=1): """Adds one classification results to the confusion matrix.""" if actual_class in self.mat: row = self.mat[actual_class] if detected_class in row: row[detected_class] += count else: row[detected_class] = count else: row = dict() row[detected_class] = count self.mat[actual_class] = row def aslistoflist(self)-
Returns the confusion matrix as list-of-lists.
Source code
def aslistoflist(self): """Returns the confusion matrix as list-of-lists.""" class_names = self._keys() n = len(class_names) idx_map = dict() for i,k in enumerate(class_names): idx_map[k] = i m = ListOfListMat(n, n) for actual_class,detected_classes in self.mat.items(): for detected_class,count in detected_classes.items(): m[ idx_map[actual_class] ][ idx_map[detected_class] ] = count return m def plot(self, title=None)-
Plots the confusion matrix.
Source code
def plot(self, title=None): """Plots the confusion matrix.""" # conf_arr = [[33,2,0,0,0,0,0,0,0,1,3], [3,31,0,0,0,0,0,0,0,0,0], [0,4,41,0,0,0,0,0,0,0,1], [0,1,0,30,0,6,0,0,0,0,1], [0,0,0,0,38,10,0,0,0,0,0], [0,0,0,3,1,39,0,0,0,0,4], [0,2,2,0,4,1,31,0,0,0,2], [0,1,0,0,0,0,0,36,0,2,0], [0,0,0,0,0,0,1,5,37,5,1], [3,0,0,0,0,0,0,0,0,39,0], [0,0,0,0,0,0,0,0,0,0,38] ] conf_arr = self.aslistoflist() class_names = self._keys() class_nums = range(0, len(class_names)) norm_conf = [] for row_i in conf_arr: row_sum = float(sum(row_i)) if row_sum > 0: normalized_row1 = [ float(j)/row_sum for j in row_i ] normalized_row = [] for j in normalized_row1: if j > 0: normalized_row.append( max(j, 0.03) ) else: normalized_row.append( 0 ) else: normalized_row = [ 0 for j in row_i ] norm_conf.append(normalized_row) #plt.clf() fig = plt.figure(figsize=(10,8)) ax = fig.add_subplot(111) mat = np.array(norm_conf) # flip upside-down mat = np.flipud(mat) class_names_r = list(reversed(class_names)) conf_arr = reversed(conf_arr) res = ax.imshow(mat, cmap=plt.cm.get_cmap("jet"), interpolation='nearest') cb = plt.colorbar(res, shrink=0.74, #pad=0.3, #extend="both", spacing="proportional", orientation='vertical') #panchor=(1.0, 0.3)) dark_color = (0,0,0) bright_color = (0.8,0.8,0.8) for i, cas in enumerate(conf_arr): for j, c in enumerate(cas): #p = norm_conf[i][j] p = mat[i,j] fcolor = dark_color if p < 0.35: fcolor = bright_color if c == 0: pass elif c >= 1 and c <= 9: plt.text(j-.20, i-.25, c, fontsize=6, color=fcolor) elif c >= 10 and c <= 99: plt.text(j-.40, i-.25, c, fontsize=6, color=fcolor) elif c >= 100 and c <= 999: plt.text(j-.40, i-.25, c, fontsize=4, color=fcolor) elif c >= 1000 and c <= 9999: plt.text(j-.40, i-.25, c, fontsize=3, color=fcolor) else: raise Exception("Too big number: "+str(c)) if title is not None: ax.set_title(title) ax.set_xlabel("classified as") ax.set_ylabel("actual class") ax.set_xlim(-0.5, max(class_nums) + 0.5) ax.set_ylim(-0.5, max(class_nums) + 0.5) import matplotlib.lines as mpllines lines = ax.get_xticklines() for line in lines: line.set_marker(mpllines.TICKDOWN) lines = ax.get_yticklines() for line in lines: line.set_marker(mpllines.TICKLEFT) plt.yticks( class_nums, [ label+" " for label in class_names_r ], rotation=0, horizontalalignment='right', fontsize=8) plt.xticks( class_nums, [ label+" " for label in class_names ], rotation=90, fontsize=8) plt.minorticks_off() plt.tick_params(direction="out", left=True, bottom=True, top=False, right=False) #plt.subplots_adjust(top=0.95, bottom=0.12, left=0.28, right=0.77) # note: # needed to re-adjust shrink,pad value as matplotlib changed its behaviour plt.subplots_adjust(top=0.95, bottom=0.12, left=0.28, right=0.90) return plt