get_performance_array() - Code Metrics - Inspection of "performance: Add performance matrix and LearningRe..." - TinghuiWang/pyActLearn - Measure and Improve Code Quality continuously with Scrutinizer

Completed

Push — master ( 23d8f0...154b9f )

by Tinghui

created 2016-11-29 03:21 UTC

get_performance_array() D

↳ Parent: Project

Complexity

Conditions

Size

Total Lines

138

Duplication

Lines	0
Ratio	0 %

Importance

Changes	1
Bugs	0	Features	0

Metric	Value
cc	8
c	1
b	0
f	0
dl	0
loc	138
rs	4

How to fix Long Method

import logging
import numpy as np

logger = logging.getLogger(__file__)

per_class_performance_index = ['true_positive', 'true_negative', 'false_positive', 'false_negative',
                               'accuracy', 'misclassification', 'recall', 'false positive rate',
                               'specificity', 'precision', 'prevalence', 'f-1 measure', 'g-measure']

overall_performance_index = ['average accuracy', 'weighed accuracy',
                             'precision (micro)', 'recall (micro)', 'f-1 score (micro)',
                             'precision (macro)', 'recall (macro)', 'f-1 score (macro)',
                             'exact matching ratio']


def get_confusion_matrix(num_classes, label, predicted):
    """Calculate confusion matrix based on ground truth and predicted result

    Args:
        num_classes (:obj:`int`): Number of classes
        label (:obj:`list` of :obj:`int`): ground truth labels
        predicted (:obj:`list` of :obj:`int`): predicted labels

    Returns:
        :class:`numpy.array`: Confusion matrix (`numpy_class` by `numpy_class`)
    """
    matrix = np.zeros((num_classes, num_classes))
    for i in range(len(label)):
        matrix[label[i]][predicted[i]] += 1
    return matrix


def get_performance_array(confusion_matrix):
    """Calculate performance matrix based on the given confusion matrix

    Gets performance array for each class
    0 - True_Positive: number of samples that belong to class and classified correctly
    1 - True_Negative: number of samples that correctly classified as not belonging to class
    2 - False_Positive: number of samples that belong to class and not classified correctMeasure:
    3 - False_Negative: number of samples that do not belong to class but classified as class
    4 - Accuracy: Overall, how often is the classifier correct? (TP + TN) / (TP + TN + FP + FN)
    5 - Misclassification: Overall, how often is it wrong? (FP + FN) / (TP + TN + FP + FN)
    6 - Recall: When it's actually yes, how often does it predict yes? TP / (TP + FN)
    7 - False Positive Rate: When it's actually no, how often does it predict yes? FP / (FP + TN)
    8 - Specificity: When it's actually no, how often does it predict no? TN / (FP + TN)
    9 - Precision: When it predicts yes, how often is it correct? TP / (TP + FP)
    10 - Prevalence: How often does the yes condition actually occur in our sample? Total(class) / Total(samples)
    11 - F(1) Measure: 2 * (precision * recall) / (precision + recall)
    12 - G Measure:  sqrt(precision * recall)

    Gets Overall Performance for the classifier
    0 - Average Accuracy: The average per-class effectiveness of a classifier
    1 - Weighed Accuracy: The average effectiveness of a classifier weighed by prevalence of each class
    2 - Precision (micro): Agreement of the class labels with those of a classifiers if calculated from sums of per-text
                           decision
    3 - Recall (micro): Effectiveness of a classifier to identify class labels if calculated from sums of per-text
                        decisions
    4 - F-Score (micro): Relationship between data's positive labels and those given by a classifier based on a sums of
                         per-text decisions
    5 - Precision (macro): An average per-class agreement of the data class labels with those of a classifiers
    6 - Recall (macro): An average per-class effectiveness of a classifier to identify class labels
    7 - F-Score (micro): Relations between data's positive labels and those given by a classifier based on a per-class
                         average
    8 - Exact Matching Ratio: The average per-text exact classification

    Note: In Multi-class classification, Micro-Precision == Micro-Recall == Micro-FScore == Exact Matching Ratio
    (Multi-class classification: each input is to be classified into one and only one class)

    Reference Document:
    Sokolova, M., & Lapalme, G. (2009). A systematic analysis of performance measures for classification tasks.
    Information Processing and Management, 45, p. 427-437

    Args:
        num_classes (:obj:`int`): Number of classes
        confusion_matrix (:class:`numpy.array`): Confusion Matrix (numpy array of num_class by num_class)

    Returns:
        :obj:`tuple` of :class:`numpy.array`: tuple of overall performance and per class performance
    """
    if confusion_matrix.shape[0] != confusion_matrix.shape[1]:
        logger.error("confusion matrix with shape " + str(confusion_matrix.shape) + " is not square.")
        return None, None

    num_classes = confusion_matrix.shape[0]

    per_class = np.zeros((num_classes, len(per_class_performance_index)), dtype=float)
    overall = np.zeros((len(overall_performance_index),), dtype=float)

    for i in range(num_classes):
        true_positive = confusion_matrix[i][i]
        true_negative = np.sum(confusion_matrix)\
            - np.sum(confusion_matrix[i, :])\
            - np.sum(confusion_matrix[:, i])\
            + confusion_matrix[i][i]
        false_positive = np.sum(confusion_matrix[:, i]) - confusion_matrix[i][i]
        false_negative = np.sum(confusion_matrix[i, :]) - confusion_matrix[i][i]
        # Accuracy: (TP + TN) / (TP + TN + FP + FN)
        per_class_accuracy = (true_positive + true_negative)\
            / (true_positive + true_negative + false_positive + false_negative)
        # Mis-classification: (FP + FN) / (TP + TN + FP + FN)
        per_class_misclassification = (false_positive + false_negative)\
            / (true_positive + true_negative + false_positive + false_negative)
        # Recall: TP / (TP + FN)
        if true_positive + false_negative == 0:
            per_class_recall = 0.
        else:
            per_class_recall = true_positive / (true_positive + false_negative)
        # False Positive Rate: FP / (FP + TN)
        if false_positive + true_negative == 0:
            per_class_fpr = 0.
        else:
            per_class_fpr = false_positive / (false_positive + true_negative)
        # Specificity: TN / (FP + TN)
        if false_positive + true_negative == 0:
            per_class_specificity = 0.
        else:
            per_class_specificity = true_negative / (false_positive + true_negative)
        # Precision: TP / (TP + FP)
        if true_positive + false_positive == 0:
            per_class_precision = 0.
        else:
            per_class_precision = true_positive / (true_positive + false_positive)
        # prevalence
        per_class_prevalence = (true_positive + false_negative)\
            / (true_positive + true_negative + false_positive + false_negative)
        # F-1 Measure: 2 * (precision * recall) / (precision +
        if per_class_precision + per_class_recall == 0:
            per_class_fscore = 0.
        else:
            per_class_fscore = 2 * (per_class_precision * per_class_recall) / (per_class_precision + per_class_recall)
        # G Measure: sqrt(precision * recall)
        per_class_gscore = np.sqrt(per_class_precision * per_class_recall)
        per_class[i][0] = true_positive
        per_class[i][1] = true_negative
        per_class[i][2] = false_positive
        per_class[i][3] = false_negative
        per_class[i][4] = per_class_accuracy
        per_class[i][5] = per_class_misclassification
        per_class[i][6] = per_class_recall
        per_class[i][7] = per_class_fpr
        per_class[i][8] = per_class_specificity
        per_class[i][9] = per_class_precision
        per_class[i][10] = per_class_prevalence
        per_class[i][11] = per_class_fscore
        per_class[i][12] = per_class_gscore

    # Average Accuracy: Sum{i}{Accuracy{i}} / num_class
    overall[0] = np.sum(per_class[:, per_class_performance_index.index('accuracy')]) / num_classes
    # Weighed Accuracy: Sum{i}{Accuracy{i} * Prevalence{i}} / num_class
    overall[1] = np.dot(per_class[:, per_class_performance_index.index('accuracy')],
                        per_class[:, per_class_performance_index.index('prevalence')])
    # Precision (micro): Sum{i}{TP_i} / Sum{i}{TP_i + FP_i}
    overall[2] = np.sum(per_class[:, per_class_performance_index.index('true_positive')]) / \
                 np.sum(per_class[:, per_class_performance_index.index('true_positive')] +
                        per_class[:, per_class_performance_index.index('false_positive')])
    # Recall (micro): Sum{i}{TP_i} / Sum{i}{TP_i + FN_i}
    overall[3] = np.sum(per_class[:, per_class_performance_index.index('true_positive')]) / \
                 np.sum(per_class[:, per_class_performance_index.index('true_positive')] +
                        per_class[:, per_class_performance_index.index('false_negative')])
    # F_Score (micro): 2 * Precision_micro * Recall_micro / (Precision_micro + Recall_micro)
    overall[4] = 2 * overall[2] * overall[3] / (overall[2] + overall[3])
    # Precision (macro): Sum{i}{Precision_i} / num_class
    overall[5] = np.sum(per_class[:, per_class_performance_index.index('precision')]) / num_classes
    # Recall (macro): Sum{i}{Recall_i} / num_class
    overall[6] = np.sum(per_class[:, per_class_performance_index.index('recall')]) / num_classes
    # F_Score (macro): 2 * Precision_macro * Recall_macro / (Precision_macro + Recall_macro)
    overall[7] = 2 * overall[5] * overall[6] / (overall[5] + overall[6])
    # Exact Matching Ratio:
    overall[8] = np.trace(confusion_matrix) / np.sum(confusion_matrix)
    return overall, per_class



1			import logging
2			import numpy as np
3
4			logger = logging.getLogger(__file__)
5
6			per_class_performance_index = ['true_positive', 'true_negative', 'false_positive', 'false_negative',
7			'accuracy', 'misclassification', 'recall', 'false positive rate',
8			'specificity', 'precision', 'prevalence', 'f-1 measure', 'g-measure']
9
10			overall_performance_index = ['average accuracy', 'weighed accuracy',
11			'precision (micro)', 'recall (micro)', 'f-1 score (micro)',
12			'precision (macro)', 'recall (macro)', 'f-1 score (macro)',
13			'exact matching ratio']
14
15
16			def get_confusion_matrix(num_classes, label, predicted):
17			"""Calculate confusion matrix based on ground truth and predicted result
18
19			Args:
20			num_classes (:obj:`int`): Number of classes
21			label (:obj:`list` of :obj:`int`): ground truth labels
22			predicted (:obj:`list` of :obj:`int`): predicted labels
23
24			Returns:
25			:class:`numpy.array`: Confusion matrix (`numpy_class` by `numpy_class`)
26			"""
27			matrix = np.zeros((num_classes, num_classes))
28			for i in range(len(label)):
29			matrix[label[i]][predicted[i]] += 1
30			return matrix
31
32
33			def get_performance_array(confusion_matrix):
34			"""Calculate performance matrix based on the given confusion matrix
35
36			Gets performance array for each class
37			0 - True_Positive: number of samples that belong to class and classified correctly
38			1 - True_Negative: number of samples that correctly classified as not belonging to class
39			2 - False_Positive: number of samples that belong to class and not classified correctMeasure:
40			3 - False_Negative: number of samples that do not belong to class but classified as class
41			4 - Accuracy: Overall, how often is the classifier correct? (TP + TN) / (TP + TN + FP + FN)
42			5 - Misclassification: Overall, how often is it wrong? (FP + FN) / (TP + TN + FP + FN)
43			6 - Recall: When it's actually yes, how often does it predict yes? TP / (TP + FN)
44			7 - False Positive Rate: When it's actually no, how often does it predict yes? FP / (FP + TN)
45			8 - Specificity: When it's actually no, how often does it predict no? TN / (FP + TN)
46			9 - Precision: When it predicts yes, how often is it correct? TP / (TP + FP)
47			10 - Prevalence: How often does the yes condition actually occur in our sample? Total(class) / Total(samples)
48			11 - F(1) Measure: 2 * (precision * recall) / (precision + recall)
49			12 - G Measure: sqrt(precision * recall)
50
51			Gets Overall Performance for the classifier
52			0 - Average Accuracy: The average per-class effectiveness of a classifier
53			1 - Weighed Accuracy: The average effectiveness of a classifier weighed by prevalence of each class
54			2 - Precision (micro): Agreement of the class labels with those of a classifiers if calculated from sums of per-text
55			decision
56			3 - Recall (micro): Effectiveness of a classifier to identify class labels if calculated from sums of per-text
57			decisions
58			4 - F-Score (micro): Relationship between data's positive labels and those given by a classifier based on a sums of
59			per-text decisions
60			5 - Precision (macro): An average per-class agreement of the data class labels with those of a classifiers
61			6 - Recall (macro): An average per-class effectiveness of a classifier to identify class labels
62			7 - F-Score (micro): Relations between data's positive labels and those given by a classifier based on a per-class
63			average
64			8 - Exact Matching Ratio: The average per-text exact classification
65
66			Note: In Multi-class classification, Micro-Precision == Micro-Recall == Micro-FScore == Exact Matching Ratio
67			(Multi-class classification: each input is to be classified into one and only one class)
68
69			Reference Document:
70			Sokolova, M., & Lapalme, G. (2009). A systematic analysis of performance measures for classification tasks.
71			Information Processing and Management, 45, p. 427-437
72
73			Args:
74			num_classes (:obj:`int`): Number of classes
75			confusion_matrix (:class:`numpy.array`): Confusion Matrix (numpy array of num_class by num_class)
76
77			Returns:
78			:obj:`tuple` of :class:`numpy.array`: tuple of overall performance and per class performance
79			"""
80			if confusion_matrix.shape[0] != confusion_matrix.shape[1]:
81			logger.error("confusion matrix with shape " + str(confusion_matrix.shape) + " is not square.")
82			return None, None
83
84			num_classes = confusion_matrix.shape[0]
85
86			per_class = np.zeros((num_classes, len(per_class_performance_index)), dtype=float)
87			overall = np.zeros((len(overall_performance_index),), dtype=float)
88
89			for i in range(num_classes):
90			true_positive = confusion_matrix[i][i]
91			true_negative = np.sum(confusion_matrix)\
92			- np.sum(confusion_matrix[i, :])\
93			- np.sum(confusion_matrix[:, i])\
94			+ confusion_matrix[i][i]
95			false_positive = np.sum(confusion_matrix[:, i]) - confusion_matrix[i][i]
96			false_negative = np.sum(confusion_matrix[i, :]) - confusion_matrix[i][i]
97			# Accuracy: (TP + TN) / (TP + TN + FP + FN)
98			per_class_accuracy = (true_positive + true_negative)\
99			/ (true_positive + true_negative + false_positive + false_negative)
100			# Mis-classification: (FP + FN) / (TP + TN + FP + FN)
101			per_class_misclassification = (false_positive + false_negative)\
102			/ (true_positive + true_negative + false_positive + false_negative)
103			# Recall: TP / (TP + FN)
104			if true_positive + false_negative == 0:
105			per_class_recall = 0.
106			else:
107			per_class_recall = true_positive / (true_positive + false_negative)
108			# False Positive Rate: FP / (FP + TN)
109			if false_positive + true_negative == 0:
110			per_class_fpr = 0.
111			else:
112			per_class_fpr = false_positive / (false_positive + true_negative)
113			# Specificity: TN / (FP + TN)
114			if false_positive + true_negative == 0:
115			per_class_specificity = 0.
116			else:
117			per_class_specificity = true_negative / (false_positive + true_negative)
118			# Precision: TP / (TP + FP)
119			if true_positive + false_positive == 0:
120			per_class_precision = 0.
121			else:
122			per_class_precision = true_positive / (true_positive + false_positive)
123			# prevalence
124			per_class_prevalence = (true_positive + false_negative)\
125			/ (true_positive + true_negative + false_positive + false_negative)
126			# F-1 Measure: 2 * (precision * recall) / (precision +
127			if per_class_precision + per_class_recall == 0:
128			per_class_fscore = 0.
129			else:
130			per_class_fscore = 2 * (per_class_precision * per_class_recall) / (per_class_precision + per_class_recall)
131			# G Measure: sqrt(precision * recall)
132			per_class_gscore = np.sqrt(per_class_precision * per_class_recall)
133			per_class[i][0] = true_positive
134			per_class[i][1] = true_negative
135			per_class[i][2] = false_positive
136			per_class[i][3] = false_negative
137			per_class[i][4] = per_class_accuracy
138			per_class[i][5] = per_class_misclassification
139			per_class[i][6] = per_class_recall
140			per_class[i][7] = per_class_fpr
141			per_class[i][8] = per_class_specificity
142			per_class[i][9] = per_class_precision
143			per_class[i][10] = per_class_prevalence
144			per_class[i][11] = per_class_fscore
145			per_class[i][12] = per_class_gscore
146
147			# Average Accuracy: Sum{i}{Accuracy{i}} / num_class
148			overall[0] = np.sum(per_class[:, per_class_performance_index.index('accuracy')]) / num_classes
149			# Weighed Accuracy: Sum{i}{Accuracy{i} * Prevalence{i}} / num_class
150			overall[1] = np.dot(per_class[:, per_class_performance_index.index('accuracy')],
151			per_class[:, per_class_performance_index.index('prevalence')])
152			# Precision (micro): Sum{i}{TP_i} / Sum{i}{TP_i + FP_i}
153			overall[2] = np.sum(per_class[:, per_class_performance_index.index('true_positive')]) / \
154			np.sum(per_class[:, per_class_performance_index.index('true_positive')] +
155			per_class[:, per_class_performance_index.index('false_positive')])
156			# Recall (micro): Sum{i}{TP_i} / Sum{i}{TP_i + FN_i}
157			overall[3] = np.sum(per_class[:, per_class_performance_index.index('true_positive')]) / \
158			np.sum(per_class[:, per_class_performance_index.index('true_positive')] +
159			per_class[:, per_class_performance_index.index('false_negative')])
160			# F_Score (micro): 2 * Precision_micro * Recall_micro / (Precision_micro + Recall_micro)
161			overall[4] = 2 * overall[2] * overall[3] / (overall[2] + overall[3])
162			# Precision (macro): Sum{i}{Precision_i} / num_class
163			overall[5] = np.sum(per_class[:, per_class_performance_index.index('precision')]) / num_classes
164			# Recall (macro): Sum{i}{Recall_i} / num_class
165			overall[6] = np.sum(per_class[:, per_class_performance_index.index('recall')]) / num_classes
166			# F_Score (macro): 2 * Precision_macro * Recall_macro / (Precision_macro + Recall_macro)
167			overall[7] = 2 * overall[5] * overall[6] / (overall[5] + overall[6])
168			# Exact Matching Ratio:
169			overall[8] = np.trace(confusion_matrix) / np.sum(confusion_matrix)
170			return overall, per_class
171
172

TinghuiWang / pyActLearn

Push — master ( 23d8f0...154b9f )

get_performance_array() D

Complexity

Size

Duplication

Importance

How to fix Long Method

Long Method

Duplication Side-by-Side

Filter issues like