A major step in most predictive analytics workflows is to create features from input data that can be fed into machine learning algorithms. This is often a manual and labor-intensive effort. Feature learning (also known as representation learning) allows important features to be automatically extracted from raw input data.
Topics that are covered:
- Manual feature engineering vs. feature learning
- Example applications of feature learning
- Matrix factorization approaches (deep dive into PCA/SVD)
- Neural network approaches (deep dive into Autoencoders and Skip-Gram/Word2Vec)
- Code samples using scikit-learn and keras
The make-data.ipynb does NOT need to be run, the feature-learning.ipynb pulls a pre-processed dataset from S3.
Slides: https://github.com/rikturr/mml-feature-learning/blob/master/slides.pdf