ledovsky / kaggle-allstate-solution

My solution of Kaggle competition - Allstate Claims Severity

Allstate - Claims Severity. Top 5-% solution

I've uploaded this solution, generally, for myself and also to be able to show it to my friends. Although, if you have any questions related to this repository, don't hesitate to contact me.

Competition link

What you can find here

• XGBoost gradient boosting
• LightGBM gradient boosting
• Keras neural networks
• HyperOpt for parameter tuning
• Stacking with linear model

What to see

• First level models and their tuning can be found in scripts
• notebooks has some experiments and also second level model (4 - Stacking.ipynb)
• Files that are used for final submission marked as "final"

Interesting public scripts

• Dmitry Dremov starting notebook with stacking of XGBoost и LightGBM
• Iglovikov - XGBoost 1114
• Best public XGBoost (1106.3) with feature extraction
• Keras - bagging - 1111
• Markov Chain Monte Carlo example
• Bayesian optimization example

Useful links

• Xgboost hyperparameter optimization notes
• Stacking strategies

Solution explanation from winners

• 1st place - BishwarupB
• 2nd place - Nikolay Noskov
• 3rd place - Faron