ML Evaluation Cheatsheet

Classification

\(FNR = \frac{F_n}{F_n + T_p}\) (False Negative Rate)

\(FPR = \frac{F_p}{F_p + T_n}\) (False Positive Rate)

\(TPR = \frac{T_p}{T_p + F_n}\) (True Positive Rate)

\(TNR = \frac{T_n}{T_n + F_p}\) (True Negative Rate)

\(P = \frac{T_p}{T_p + F_p}\) (Precision)

\(R = \frac{T_p}{T_p + F_n}\) (Recall)

\(F_1 = 2\frac{P \times R}{P + R}\) (F1 Score)

\(ACC = \frac{T_p + T_n}{T_p + T_n + F_p + F_n}\) (Accuracy)

Landmark Detection

\(NME_k = \frac{1}{N_L} \sum_{i=1}^{N_L} \frac{\text{\textbardbl} p_i - \^p_i \text{\textbardbl}_2 }{d} \) (Normalized Mean Error)

\(NME = \frac{1}{K} \sum_{k=1}^{K} NME_k \) (Mean Normalized Mean Error)

\(FR = \frac{1}{K} \sum_{k=1}^{K} [NME_k \ge Threshold]\) (Failure Rate)

* \(K = \) Number of samples

* \(N_L = \) Number landmarks on sample

References