Post by Tiago Albineli Motta

"The model looked promising offline. Which, as every machine learning practitioner knows, is usually the beginning of the story, not the end. The model performed well in validation. Production disagreed. Fortunately, this wasn’t caused by overfitting, data leakage, or somebody accidentally training on the target variable while pretending not to. The issue was simpler: the training population and the inference population were fundamentally different." Amazing work with the dearly missed Profile & Segmentation team at Globo.com: Bruno de F. Melo e Souza Paulo Azize Diogo Munaro Vieira Arnour Sabino Gabriel Faleiro Fernanda Rabelo Paola Benchimol https://lnkd.in/dM7622Bk