2. Goals
● Scenario-1: In a course project or competition with given evaluation scheme
and metric, I want to understand the metric to do achieve better result.
● Scenario-2: In a practical project, I need to choose an evaluation scheme and
metric that fit the model and the business need.
3. Popular metrics in Kaggle competitions
Data source: https://www.kaggle.com/kaggle/meta-kaggle
4. Popular metrics in Kaggle competitions
Data source: https://www.kaggle.com/kaggle/meta-kaggle
5. Popular metrics in Kaggle competitions
Data source: https://www.kaggle.com/kaggle/meta-kaggle
6. Regression Metrics
● Root Mean Squared Error
● Mean Absolute Error
● Root Mean Squared Logarithmic Error
14. Examples using Root Mean Squared Error
● Relevance score of a product search result chosen from 1, 2, 3 (is RMSE and
MAE the same in this case?)
● A 0-100 rating of a person for a piece of music
● Travel time on highway
● Probabilities of a galaxy belonging to a certain class
15. Examples using Mean Absolute Error
● How much loan in $ defaulted
● Certain part on computers of a certain model, how many repairs are required
each month
● Tens of target variables transformed to be on the same scale, meanings
unknown
● Salary of a job
● Solar radiation
● ...
16. Regression Metrics
● Root Mean Squared Error
● Mean Absolute Error
● Root Mean Squared Logarithmic Error
20. ● Days in hospital a patient will spend in a year
● Number of edits a person will make on Wikipedia in 5 months
● Online sales for a product in a year
● Bulldozer prices
Examples using Root Mean Squared Logarithmic Error
21. ● Days in hospital a patient will spend in a year
● Number of edits a person will make on Wikipedia in 5 months
● Online sales for a product in a year
● Bulldozer prices
Examples using Root Mean Squared Logarithmic Error
41. Examples using Log Loss
● Does an insurance claim need additional information
● Game outcomes in March Madness
● Will a user click an add
● Image classification tasks
42. Multilabel Classification
● Average of the AUCs of each label
● MAP@K
E.g., Which categories does an online comment belong to
(toxic,severe_toxic,obscene,threat,insult,identity_hate)