# 🎖️ 1st Rainfall binary prediction

👋 Hi, everyone! Now i give more information about my solution:
[TOP 1 Solution Notebook](https://www.kaggle.com/code/antonoof/top1-tabpfn-rainfall-binary-prediction)
In this notebook i use and improve code create by
[@philippsinger](https://www.kaggle.com/philippsinger)
, Thank you for your solution! 🙏
TabPFN (Tabular Prior-Data Fitted Network) is a pre—trained Transformer architecture model designed specifically for working with small tabular data.
Wait, Transformer? It's for texts and pictures! Exactly. The PriorLabs developers took the same idea that underlies GPT and other large language models and applied it to tables. They generated millions of synthetic datasets and trained a model on them. As a result, TabPFN has "learned to learn" from tabular data.
When you give him your dataset (up to 50,000 lines in size), he doesn't learn from scratch. Instead, he uses his accumulated "knowledge" about the structures and dependencies in the tables to instantly make a prediction. This process is called "In-context Learning", and it is thanks to it that the model works so quickly and does not require configuration.
![TabPFN](assets/4fcde2471a8a659f02465cbc.webp)