The challenge is decribed in: https://www.kaggle.com/c/titanic.
Data samples are available on the competition web page.
There are several notebooks provided:
- EDA + feature extraction (used to create derived datasets)
- detailed comparison of LightGBM vs XGBoost with different preprocessing (an extensive performance comparison using train/valid split as well as cross validation with various models and feature transformers)
- basic comparison of LightGBM vs XGBoost (a bsaic comparison of different models. This is also available as a kaggle kernel: https://www.kaggle.com/mlisovyi/lightgbm-vs-xgboost-sklearn-api)
- Hyper-parameter optimisation of XGBoost and LightGBM models
- Example of linear model based on text features using Vowpal Wabbit (This is also available as a kaggle kernel: https://www.kaggle.com/mlisovyi/vowpal-wabbit-decides-who-lives-and-who-dies)
- Submission preparation
If notebooks are not rendered, when you open then from github in your browser, you can view them using https://nbviewer.jupyter.org/
The notebooks are developed in python3.5, but most of the code will most likely work in any python3 environment. All of those can be installed with conda (either mini- or ana-). Just do something like
conda install numpy pandas matplotlib seaborn scikit-learn ipykernel jupyter xgboost lightgbmYou might need to add conda-forge to the list of channels: conda config --add channels conda-forge,
if you have not done so yet (the installation would fail complaining that a subset of modules can not be found).
If you want to try the Vowpal Wabbit (https://github.com/JohnLangford/vowpal_wabbit) for classification, you will need to play additional tricks: read and follow the instructions on their README of the python package. In practice, on a Linux system with conda installed you will need to:
- use a conda environment with python3.5 or python3.6;
conda install -c anaconda boostthis is important, as it installs a series of boost packages, that are compatible with your conda environment (i.e. python version);pip install vowpalwabbitintall the python binding + the cython library of the VW itself. This procedure has been tested on Ubuntu 16.04 and 18.04 with a python3.5 and python3.6 environments in conda and geting VW from github.