What this book covers
Chapter 1, The Most Renowned Tabular Competition – Porto Seguro’s Safe Driver Prediction. In this competition, you are asked to solve a common problem in insurance to figure out who is going to have an auto insurance claim in the next year. We guide you in properly using LightGBM, denoising autoencoders, and how to effectively blend them.
Chapter 2, The Makridakis Competitions – M5 on Kaggle for Accuracy and Uncertainty. In this competition based on Walmart’s daily sales time series of items hierarchically arranged into departments, categories, and stores spread across three U.S. states, we recreate the 4th-place solution’s ideas from Monsaraida to demonstrate how we can effectively use LightGBM for this time series problem.
Chapter 3, Vision Competition – Cassava Leaf Disease Classification. In this contest, the participants were tasked with classifying crowdsourced photos of cassava plants grown by farmers in Uganda. We use the multiclass problem to demonstrate how to build a complete pipeline for image classification and show how this baseline can be utilized to construct a competitive solution using a vast array of possible extensions.
Chapter 4, NLP Competition – Google Quest Q&A Labeling, discusses a contest focused on predicting human responders’ evaluations of subjective aspects of a question-answer pair, where an understanding of context was crucial. Casting the challenge as a multiclass classification problem, we build a baseline solution exploring the semantic characteristics of a corpus, followed by an examination of more advanced methods that were necessary for leaderboard ascent.