cm2metrics

A lightweight package that analyzes multiple metrics directly from confusion matrix.

Features

• Get all classes parsing results in native Dataframe format
• Print all classes or specified class parsing summary in friendly format
• Only requires numpy and pandas, without relying on other machine learning packages.
• Easy to use with a few APIs
• Supports 16 metrics for each class:
  • tp: true positive
  • tn: true negative
  • fp: false positive
  • fn: false negative
  • tpr: true positive rate
  • tnr: true negative rate
  • fpr: false positive rate
  • fnr: false negative rate
  • atc: actual true count
  • afc: actual false count
  • ptc: predict true count
  • pfc: predict false count
  • accruacy
  • precision
  • recall
  • f1

General

• Version: 0.1
• Dependency: Python(3.6,3.7.3.8), numpy, pandas

Install

pip install cm2metrics

Use

General use

Generate a confusion matrix

#  use scikitlearn
from sklearn.metrics import confusion_matrix

#cm is ndarray, convert to dataframe
cm = confusion_matrix(true_target, pred_target)
df_cm = pd.DataFrame(cm, index=class_names, columns=class_names)

# or, use a randomly generated confusion matrix(for test)
# see details in cm_test.py
class_names = {0:"class0", 1:"class1", 2:"class2"}
df_cm = pd.DataFrame([(1,2,3),(4,5,6),(7,8,9)])
df_cm.rename(index=class_names, columns=class_names, inplace=True)

Init a confusion matrix parser

from cm2metrics.parse_cm import  ConfusionMatrixParser
cm_parser = ConfusionMatrixParser(df_cm)

Parse the confusion matrix

# parsing result(cm_parsed) is a dataframe
cm_parsed = cm_parser.parse_confusion_matrix()
print(cm_parsed)  

Sample output:

        tp  fp  tn  fn       tpr       fpr       tnr       fnr  atc  afc  ptc  pfc  accuracy  precision    recall        f1
class0   1  28  11   5  0.166667  0.717949  0.282051  0.833333    6   39   12   33  0.644444   0.083333  0.166667  0.111111
class1   5  20  10  10  0.333333  0.666667  0.333333  0.666667   15   30   15   30  0.555556   0.333333  0.333333  0.333333
class2   9  12   9  15  0.375000  0.571429  0.428571  0.625000   24   21   18   27  0.466667   0.500000  0.375000  0.428571


# get class0 true positive
tp = cm_parsed.loc["class0"].at["tp"]

Print parsing summary in friendly format

# print one class summary by name using class_name parameter
cm_parser.print_summary(class_name="class0")

# print one class summary by index in confusion matrix using class_index parameter
cm_parser.print_summary(class_index=0)

# print all classes summary by not specifying parameters
cm_parser.print_summary()

Sample output for class0 summary:
Summary for class0
TP: 1
TN: 28
FP: 11
FN: 5
TPR: 0.167
TNR: 0.718
FPR: 0.282
FNR: 0.833
Actual true count: 6
Actual false count: 39
Predict true count: 12
Predict false count: 33
Accuracy: 0.644
Precision: 0.083
Recall: 0.167
F1: 0.111

License

MIT license.