calculate the accuracy, precision, recall and specificity from multi-class confusion matrix

classification report.py

#import torch
import os
import numpy as np

n_classes = 5
cm = np.array
      ([[10606.4167,  1464.2083,    61.9167,    11.75  ,   301.4167],
       [  610.5417,  2315.7917,   686.4167,    59.4583,   768.9167],
       [  152.7083,  2469.5833,  8752.2083,  1394.5   ,  1248.5833],
       [    3.3333,    29.4167,   385.0833,  1920.375 ,    24.5   ],
       [  211.9167,  1287.5833,   614.7083,    71.875 ,  2945.3333]])

# np.set_printoptions(suppress=True, precision=4)


for c in range(n_classes):
    tp = cm[c,c]
    fp = sum(cm[:,c]) - cm[c,c]
    fn = sum(cm[c,:]) - cm[c,c]
    tn = sum(np.delete(sum(cm)-cm[c,:],c))

    recall = tp/(tp+fn)
    precision = tp/(tp+fp)
    specificity = tn/(tn+fp)
    f1_score = 2*((precision*recall)/(precision+recall))
    

    #print(f"for class {c}: acc {accuracy}, recall {recall},\
    #      precision {precision}, f1 {f1_score}")
    print("for class {}: recall {}, specificity {}\
          precision {}, f1 {}".format(c,round(recall,4), round(specificity,4), round(precision,4),round(f1_score,4)))

##    print("tp: ", tp)
##    print("fp: ", fp)
##    print("fn: ", fn)
##    print("tn: ", tn)