{"slug": "when-qini-rewards-the-wrong-uplift-model-use-doubly-robust-scores", "title": "When Qini Rewards the Wrong Uplift Model, Use Doubly Robust Scores", "summary": "On 1 million simulated customers, calibrated Qini scored a Linear DR learner at 0.0856 versus 0.0826 for the true-CATE ranking that should have won, according to a Towards Data Science analysis arguing that doubly robust scores should replace Qini when evaluating uplift models. The finding indicates that Qini can reward the wrong uplift model, making doubly robust scoring the more reliable evaluation choice.", "body_md": "# When Qini Rewards the Wrong Uplift Model, Use Doubly Robust Scores\n\nOn 1 million simulated customers, calibrated Qini scored a Linear DR learner 0.0856 against 0.0826 for the true-CATE ranking that should have won. The post When Qini Rewards the Wrong Uplift Model, Use Doubly Robust Scores appeared first on Towards Data Science.\n\n## Key Takeaways\n\n- •Latest on When Qini Rewards the Wrong Uplift Model, Use Doubly Robust Scores — reported by Towards Data Science for the newsletter space.\n- •This story was reported by **Towards Data Science** , covering developments in the**newsletter** space.\n- •AI advancements continue to reshape industries — read the full article on Towards Data Science for complete coverage.\n\n📖 Continue reading the full article:\n\n[Read Full Article on Towards Data Science →](https://towardsdatascience.com/when-qini-rewards-the-wrong-uplift-model-use-doubly-robust-scores/)", "url": "https://wpnews.pro/news/when-qini-rewards-the-wrong-uplift-model-use-doubly-robust-scores", "canonical_source": "https://ainexusdaily.vercel.app/article/2026-11-18-when-qini-rewards-the-wrong-uplift-model-use-doubly-robust-scores", "published_at": "2026-11-18 13:30:00+00:00", "updated_at": "2026-11-18 16:14:20.089594+00:00", "lang": "en", "topics": ["machine-learning", "ai-research"], "entities": ["Towards Data Science", "Qini", "Linear DR learner"], "also_reported_by": [], "takeaway": "Calibrated Qini scored a Linear DR learner 0.0856 against 0.0826 for the true-CATE ranking on 1 million simulated customers, so use doubly robust scores instead of Qini when the metric can reward the wrong uplift model.", "recommendation": {"advice": "When comparing uplift models, score them with doubly robust metrics rather than calibrated Qini, which on 1 million simulated customers ranked a Linear DR learner (0.0856) above the true-CATE ranking (0.0826).", "playbook": "evals-testing", "playbook_url": "https://wpnews.pro/playbooks/evals-testing"}, "alternates": {"html": "https://wpnews.pro/news/when-qini-rewards-the-wrong-uplift-model-use-doubly-robust-scores", "markdown": "https://wpnews.pro/news/when-qini-rewards-the-wrong-uplift-model-use-doubly-robust-scores.md", "text": "https://wpnews.pro/news/when-qini-rewards-the-wrong-uplift-model-use-doubly-robust-scores.txt", "jsonld": "https://wpnews.pro/news/when-qini-rewards-the-wrong-uplift-model-use-doubly-robust-scores.jsonld"}}