Sklearn make_scorer tutoriel : Créer vos propres métriques.

Item: Sklearn make_scorer tutoriel : Créer vos propres métriques.
Rating: 4
Author: PostSingularity Institute

by Guillaume Saint-Cirgue

Formal & Physical Sciences AI — Models & Research UYQArtificial intelligence (AI)UYQMMachine learning

🎙 Guillaume Saint-Cirgue 👥 204K 📅 December 11, 2019 ⏱ 13 min 👁 32K 📄 tutorial 🧭 2026-08-17

Available in: English (current) Français

Keywords

make_scorercustom metricscross-validationgrid searchscikit-learn

Summary

This tutorial by Machine Learnia explains how to create custom evaluation metrics in scikit-learn using the make_scorer function. The instructor begins by motivating the need for custom metrics in real-world projects, where clients often have specific performance criteria beyond standard metrics like R² or MSE. He illustrates this with an example where predictions must fall within a 20% tolerance of true values. He then shows how to define a custom metric function that calculates the proportion of predictions within this tolerance. The key step is wrapping this function with make_scorer, specifying whether higher scores are better. This scorer can then be used in cross-validation and grid search. The tutorial demonstrates using the custom scorer with cross_val_score and GridSearchCV to optimize an SVR model. Finally, the instructor provides an exercise for viewers to create their own RMSE scorer. The video is practical and hands-on, with code examples shown on screen.

149 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides valuable practical knowledge on a specific but important aspect of machine learning workflow: creating custom metrics. The argumentation is clear and logical, building from a real-world problem to a solution. The instructor explains not only the ‘how’ but also the ‘why’, emphasizing the importance of custom metrics in professional settings. The demonstration is effective, showing the process step-by-step and even including an error case to highlight the need for make_scorer. The tutorial is well-paced and easy to follow for viewers with basic knowledge of scikit-learn.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is adequate for a tutorial. The instructor does not cite external sources, but the content is based on standard scikit-learn documentation and practices. The tutorial is consistent with official scikit-learn usage. The title accurately describes the content. No comments were provided for analysis.

149 words

Title / Content Match

The title accurately reflects the content: a tutorial on using make_scorer to create custom metrics in sklearn.

Quality & Reliability

8/10

The tutorial is clear, well-structured, and provides a practical example. The author demonstrates expertise in the subject. However, the video lacks formal citations and references, and the code is not shown in full detail.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

Machine Learnia GitHub — Repository with code examples and resources.
Machine Learnia Website — Official website with additional tutorials and resources.
Free Book: Learn Machine Learning in One Week — Free book offered by the channel.

Concurring Sources

Scikit-learn documentation on custom scorers — Official documentation explaining how to create custom scorers.

Contribution & Novelties

The video provides a clear, practical guide to creating custom metrics in scikit-learn, a topic often overlooked in introductory tutorials. It bridges the gap between standard metrics and real-world requirements, showing how to integrate custom scorers into cross-validation and hyperparameter tuning. The tutorial is original in its focus on the make_scorer function and its application to a realistic scenario.

Pour aller plus loin :

Scikit-learn documentation on scoring — Official documentation on scoring and metrics.
Cross-validation in scikit-learn — Official guide on cross-validation.
GridSearchCV documentation — Official reference for GridSearchCV.

90 words

Radar Profile

The radar profile shows high scores in quality of information and reliability, with moderate scores in quantity and technical level. This indicates a focused, well-explained tutorial that may not cover all aspects of the topic but provides solid, reliable content.

Reliability 8/10