Performance Estimation

This page explains how to estimate performance changes in production data before ground truth (GT) is available and identify possible degradation factors.

Overview

Performance Estimation continuously monitors estimated error trends in production using existing evaluation results as a baseline. The following information is estimated in each screen.

Monitoring

Plot Area	Estimated Information
Inference Count Window	Estimated Error Count
	Estimated Critical Error Count
Moving Average Accuracy	Estimated ACC

Log Analytics

Tab	Section	Estimated Information
Error Map Tab	Error Map	Count by Estimated Error Code
	Estimated Error Codes	Category derived from Estimated Error Probability and Score
		Estimated ACC
	Estimated Error Trends	Estimated Error Code by Duration
Heatmap Tab	Delta Heatmap	Category derived from Estimated Error Probability and Score
		Difference heatmap between Error Count (with GT) and Estimated Error Count in production data (without GT)
	Count Heatmap	Category derived from Estimated Error Probability and Score
		Heatmap between Estimated Error Count in production logs (without GT) and Error Count in training data (with GT)
	Error Rate Heatmap	Category derived from Estimated Error Probability and Score
		Heatmap between Estimated Error Rate in production logs (without GT) and Error Rate in training data (with GT)