Learn Before
A Category's Share of Errors Sets an Upper Bound on Improvement
If you measure what fraction of all mistakes comes from one specific cause, that fraction is the most you could hope to remove by fixing only that cause. For example, suppose a document classifier makes 12% errors, and 25% of those errors happen on low-resolution scans. Even if low-resolution cases were solved perfectly, the total error could fall by no more than 25% of 12%, or 3 percentage points, from 12% to 9%. If instead 60% of the errors came from that one cause, the same task could potentially cut the total error from 12% to about 4.8%. The category share tells you whether that target is worth prioritizing, but it does not guarantee the full gain will be achieved.
0
1
Tags
Machine Learning
Deep Learning
Supervised Learning
Dive into Deep Learning @ D2L
Data Science
Machine Learning Strategy
Machine Learning Yearning @ DeepLearning.AI
Related
One Example May Fit Several Error Tags
New Error Categories Can Appear During Review
Choose Error Categories You Can Act On
Error Review Improves Through Repeated Passes
Using Error Counts to Decide Where to Focus Next
Working on Several Error Buckets at Once
Error Analysis Is Not an Automatic Ranking Rule
A Category's Share of Errors Sets an Upper Bound on Improvement
Error Analysis Helps Estimate Whether a Proposed Change Is Worth the Effort
Why Quick Error Review Is Often Skipped
Incorrect Labels in a Validation Set
Splitting a large development set into a review subset and a tuning subset
Build a Simple Baseline First, Then Use Error Analysis to Prioritize Improvements
Using Training-Set Mistakes to Diagnose High Bias
Reviewing a Sample of Validation Errors
Separating Search Errors from Scoring Errors in Inference
Component-Wise Error Review
Error Analysis as a Data-Science Lens on Model Mistakes
Multiple Valid Approaches to Error Analysis
Tasks Humans Can Perform Give Stronger Error Analysis Benchmarks
When diagnosing a model, what should error analysis focus on first?
Error analysis on a machine learning system must follow one fixed procedure.
Name the practice of reviewing mistakes to understand why predictions failed.
Match each error-analysis idea to the description that best fits it.
Put the steps of a simple dev-set error review in the right order.
Why is it useful to inspect misclassified examples during error analysis, even for error types you cannot immediately repair?
Error analysis is usually repeated after each round of model changes.
Error analysis can help you judge which improvement paths look most _____.
Match each error-analysis activity with the benefit it can provide.
Order the steps for deciding which error types to target after a first pass of error review.
Why Error Review Helps Set the Right Next Priorities
Plan the next review step after repeated image-classifier mistakes.
What is error analysis used for in machine learning?
Learn After
What does an error-category fraction mean as an upper limit in error analysis?
If 8% of the mistakes in a classifier come from low-light photos, then making low-light detection better will always eliminate 8% of all errors.
Category error share as a limit on improvement
Match each share of category errors to the best possible reduction in a system with an 8% error rate.
Use category fractions to judge whether an error type deserves attention
A classifier has 20% error, and 30% of its mistakes are misclassified bicycles. What is the best possible accuracy after perfectly fixing bicycle classification?
If one error category makes up a large fraction of mistakes, then fixing that category will definitely produce a large improvement in total performance.
If a model has a 12% error rate and 8% of those errors come from one specific class, the best possible error rate after fixing that class perfectly is _____.
Match each term from improvement-cap analysis with its meaning.
Order the reasoning steps for using an error-category fraction to judge priority.
Using a category’s error share to estimate its improvement limit
Which defect should a warehouse robot team fix first?
Upper bound on error reduction from one source