Applied Machine Learning

Day 1: Introduction - Start with how to define a Machine Learning use-case using an ML project canvas and learn about the industry standard lifecycle for Data Science and Machine Learning problems, CRISP-DM. We will then l provide an overview of the major steps of an end-to-end ML pipeline, including defining a base model, defining KPIs of success, different error metrics, model assessment/selection, train-test split, data cleaning, and transformation pipelines, under vs. over-fitting, performance metrics for regression tasks. Day 2: End-to-End ML pipeline - Building upon concepts from day 1, you will apply feature normalization, missing value imputation, scaling, and selection pipelines. You will work with linear and polynomial regression models and regularization techniques (ridge, lasso, and elastic net).

Day 1: Classification Models - Practice data labeling techniques and applications of Machine Learning algorithms for classification tasks using logistic regression. You will also be working with concepts such as class-imbalance, precision-recall analysis (false positive vs. false negative rates), ROC curves, and the use of the confusion matrix. Day 2: Advanced Classification Models - Go deeper into Supervised Machine Learning problems using commonly used classification algorithms like K-Neighbors, Naive Bayes, Support Vector, Decision Trees and state-of-the-art Ensemble Learning techniques such as Random Forest, XGBoost and CatBoost.

Day 1: Unsupervised Learning - Along with understanding K-means, hierarchical clustering, and DBSCAN, learn how to apply these models to identify an optimal number of clusters, generate and interpret cluster labels using visualizations and statistics. Another area of focus will be applying dimensionality reduction techniques like Principal Component Analysis and t-SNE. Day 2: Machine Learning Use Case - Moving back to classification models (now with the power of unsupervised learning), you will apply your skills in a real world scenario. Another focus of this day is to see how unsupervised learning, combined with supervised learning, increases the power of ML systems.

Day 1: Model Bias and Explainability Analysis - Most Machine Learning models are biased. We will explore the steps you need to take towards making the model accountable to its results. You will apply visualization techniques to understand model bias and perform model interpretability and explainability analysis using state-of-the-art frameworks like LIME and SHAP. You will also learn about the limitations of these methods and how to identify the effect of confounding factors in such interpretations. Day 2: Hyperparameter Tuning and AUTO-ML - Though you have been developing and using your own Machine Learning pipelines, on this day, you will learn and apply how to automate most components of the Machine Learning model development and evaluation lifecycle using state-of-the-art AUTO-ML frameworks like PyCaret, TPOT and MLJAR.

Day 1: ML Model Deployment - Now that we have learned all about model development, it’s time to move on to the next step in the ML process cycle - deployment. You will get familiar with git and version control systems, learn about model deployment strategies, key steps to building and deploying a Machine Learning model on the cloud using AWS as well as locally using a virtual environment. Day 2: Experiment Tracking and Model Monitoring - Any Machine Learning model needs to be monitored after being deployed in production (even rarely used models need to be quality-checked). During the last lecture day of this course, you will apply statistical techniques to monitor model performance and quantify data drift.

Day 1: Bring your personal project and implement your complete Machine Learning pipeline. We will provide you with suggestions. Day 2: Continuation and short presentations of final projects.