Kaggle-predicting-optimal-fertilizers

This repository has my solutions to the Kaggle competition https://www.kaggle.com/competitions/playground-series-s5e6/overview

Predicting Optimal Fertilizers – Kaggle Solutions

This repository contains my work for the Kaggle competition "Predicting Optimal Fertilizers" (https://www.kaggle.com/competitions/playground-series-s5e6/overview)). It includes exploratory data analysis, feature engineering, model development, and final submissions.

🧠 Overview

The goal of this competition is to recommend three fertilizers for a given crop, soil conditions, and environmental conditions.

📁 Repository Structure

notebooks folder: contains Jupyter notebooks for EDA, modeling, and submissions scripts submissions folder: contains the csv files of predictions submitted to Kaggle

🚀 Highlights

• Stratified validation with repeated holdouts and bootstrap resampling
• XGBoost classifier tuned via randomized search
• MAP@3 optimization with ensemble averaging
• Memory-efficient feature engineering and dtype reduction

📌 Notes

• Raw data is not included due to Kaggle’s terms of use.
• To reproduce results, download the dataset from the competition page and place it in a data/ folder.

🏁 Final Score

My best public leaderboard score: 0.32931

🐳 Running the Notebooks with Docker

All notebooks in this repository were developed and executed inside the official Kaggle GPU Docker container:

• Image: gcr.io/kaggle-gpu-images/python
• Includes: Python 3.x, Jupyter, pandas, scikit-learn, XGBoost, LightGBM, and other common ML libraries

🔧 Steps to Run Locally

1. Pull the Docker image (if you haven't already):

   docker pull gcr.io/kaggle-gpu-images/python

2. Run the container with your project mounted:

   docker run -it --rm \
     -p 8888:8888 \
     -v /path/to/your/repo:/workspace \
     gcr.io/kaggle-gpu-images/python

   Replace /path/to/your/repo with the full path to your cloned GitHub repository.

3. Launch Jupyter inside the container:

   jupyter notebook --ip=0.0.0.0 --port=8888 --allow-root --NotebookApp.token=''

4. Open your browser and go to http://localhost:8888 to access the notebooks.

---

This setup ensures full compatibility with Kaggle’s environment, including GPU support and preinstalled libraries.

---

Feel free to explore the notebooks and reach out if you have questions!