Summary
In this chapter, we have discussed tabular competitions on Kaggle. Since most of the knowledge applicable in a tabular competition overlaps with standard data science knowledge and practices, we have focused our attention on techniques more specific to Kaggle.
Starting from the recently introduced Tabular Playground Series, we touched on topics relating to reproducibility, EDA, feature engineering, feature selection, target encoding, pseudo-labeling, and neural networks applied to tabular datasets.
EDA is a crucial phase if you want to get insights on how to win a competition. It is also quite unstructured and heavily dependent on the kind of data you have. Aside from giving you general advice on EDA, we brought your attention to techniques such as t-SNE and UMAP that can summarize your entire dataset at a glance. The next phase, feature engineering, is also strongly dependent on the kind of data you are working on. We therefore provided a series of possible feature...
