FFR and FAR calculating for multiclasss biometric face recognition system

I am implementing a face recognition system using facenet and svc Ml algorithm i have like 20 classes or more and I'm getting 98% accuracy im trying to calculate the FAR and FRR and the EER I'm assuming that the threshold is the probability of predicting is that correct ?

this is the code for calculating the FP,FN,TP,TN

FP = matrics.sum(axis=0) - np.diag(matrics)  
FN = matrics.sum(axis=1) - np.diag(matrics)
TP = np.diag(matrics)
TN = matrics.sum() - (FP + FN + TP)
TPR = TP/(TP+FN)
cm = ConfusionMatrix(testy, yhat_test)

from this matrix

how to calculate FAR FRR EER from these ? i tried using the probability predictions for each FP,FN,TP,TN like this

a =[]
for i in range(len(TP)):
       a.append(max(yhat_prob[TP[i]]))

print(a)

the output is :

0.6994577123013918, 0.6994577123013918, 0.4916665680599575, 
0.6994577123013918, 0.6994577123013918, 0.6994577123013918, 
0.30549183912100375, 0.30549183912100375, 0.30549183912100375, 
0.30549183912100375, 0.30549183912100375, 0.32500348178885874, 
0.30549183912100375, 0.4916665680599575, 0.30549183912100375, 
0.30549183912100375, 0.30549183912100375, 0.6994577123013918, 
0.30549183912100375]

and I don't know what to do next in case this is correct