Day 1: Core ML

Module 1: Introduction to ML

  • [Theory] Key ML concepts and terminology
  • [Theory] Possibilities and use cases
  • [Exercise] Formalizing an ML problem: credit scoring

Coffee break

Module 2: Model creation

  • [Theory] Intuitions behind learning techniques and high-level overview (nearest neighbors, linear models, logistic regression, decision trees)
  • [Exercise] Introduction to Jupyter notebooks and Python programming
  • [Exercise] Creating a predictive model with scikit-learn

Day 2: Going further with ML

Module 5: Model selection

  • [Theory] Boosting predictions’ accuracy with ensembles
  • [Exercise] Comparing models efficiently with SKLL
  • [Theory] Cross-validation

Coffee break

Module 6: Data preparation

  • [Theory] Limitations of ML
  • [Exercise] Feature engineering on select use cases (priority inbox, real-estate price prediction, credit scoring, churn detection)
  • [Exercise] Finding issues in data and fixing them in Pandas

Lunch break

Module 3: Operationalization

  • [Theory] Functioning of REST APIs and importance for ML deployment in production
  • [Exercise] Deploying your own models as scalable APIs with Microsoft Azure ML & Cortana Intelligence Suite, and querying them anywhere
  • [Theory] Overview of ML-as-a-Service and Predictive API technologies (open source, proprietary and hybrid)

Coffee break

Module 4: Evaluation

  • [Theory] Evaluating ML models: criteria, methods, basic performance measures and baselines
  • [Theory] Performance measures for classification (recall and precision)
  • [Exercise] MLaaS evaluation: Amazon ML and BigML (APIs, Python wrappers, command-line tools and web dashboards)

Lunch break

Module 7: Advanced topics: Unsupervised Learning, Deep Learning and Recommender Systems

  • [Exercise] Running automated clustering and anomaly detection with BigML’s API and interpreting results
  • [Theory] Automatic feature extraction from text and images (illustrated with Indico’s Deep Learning API)
  • [Theory] Building recommender systems by reduction to classification and collaborative filtering

Coffee break

Module 8: Developing your own use case

  • [Theory] Formulating your own problem: asking the right questions and specifying key aspects of the problem with the Machine Learning Canvas
  • [Exercise] Applying the Canvas to your own problem, or to credit scoring
  • Conclusions: recap of key take-aways