Machine-learning Starterkit
A two day course for data scientists, data analysts, and business intelligence experts interested in using Python for their day-to-day machine-learning work. The primary focus is on learning to use supervised machine-learning techniques efficiently and effectively.
1 — Getting Started
Let’s start at the beginning. We examine different types of machine-learning problems and what the available tools are. If you can not measure how well your algorithm works you can’t improve it. We will look at cross-validation, performance metrics and common mistakes. If you can not measure how well your algorithm works you can’t improve it. We will look at cross-validation, performance metrics and common mistakes.
2 — Measuring Performance
If you can not measure how well your algorithm works you can’t improve it. We will look at cross-validation, performance metrics and common mistakes. If you can not measure how well your algorithm works you can’t improve it. We will look at cross-validation, performance metrics and common mistakes. If you can not measure how well your algorithm works you can’t improve it. We will look at cross-validation, performance metrics and common mistakes.
3 — Random Forests and gradient boosting
For real world, business data it is hard to beat random forest or gradient boosted tree models. They regularly take top spot in Kaggle competitions. In this module we will dive into tree based models. Their strengths, weaknesses and some tricks of the trade to make them work really well.
4 — Neural networks and deep learning
Neural networks are the foundation of the current deep learning and AI resurgence. They rule supreme when it comes to image classification and object detection. In addition most state of the art models for sound and natural language understanding are based on neural networks. We will venture into the field of deep learning and its applications to image classification, object detection and image segmentation.
5 — Interpretable decisions and debugging
Most machine-learning are seen as black-box decision makers. In large part because the tools and techniques for looking inside are not as well known as they should be. In this module we will learn about techniques like LIME and partial dependence plots. We will apply them to neural networks and tree based ensembles to debug our models and generate explanations for customers.
6 — Closing
Recap of what you learned and pointers to resources to keep learning. Stay afterwards for an informal chat and networking.