This is the code for models that ended-up in top 10 in Kaggle's 'Dato - Truly Native?' contest

https://www.kaggle.com/c/dato-native

This code mainly builds XGBoost and MLP Keras models using HashingVectorizer (http://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.HashingVectorizer.html) features. Here are the steps to build the models:

• Assume that all data (i.e. all csv and html_txt files) is present in the data/ directory.
• Build HashingVectorizer features first using src/FeatureExtractor.py
• Learn a slightly-weak Naive-Bayes Classifier using all features
• Perform feature-extraction using Naive-Bayes feature importance
• Convert those important features in LibSVM format for easy sparse encoding
• Learn both XGBoost and Keras models using these features

We have provided a simple run.sh script automate these steps. An ensemble of these two models alone could give you a validation AUC-score of close to 0.984