7 May 2019
Franz Kiraly - Supervised learning with skpro
A hands-on session in building and assessing predictive models that also produce uncertainty estimates of a their prediction, such as: predictive intervals, and fully probabilistic supervised predictions including Bayesian predictive posteriors. Starting with a gentle theoretical overview introduction on what are, how to make, and how to evaluate probabilistic classification/regression models, followed by a demo of the skpro package, "scikit-learn for probabilistic supervised learning".
The session will also feature a short overview of python modelling toolboxes currently in development at the Alan Turing Institute, and how to get involved.
Jim Dowling - The Feature Store: how Python connects Data Engineering with Data Science in Scalable ML Pipelines
Machine Learning (ML) pipelines are the fundamental building block for productionizing ML code. However, existing tutorials and educational material in Python for Data Scientists emphasizes ad-hoc feature engineering and training pipelines to experiment with ML models. Such pipelines have a tendency to become complex over time and do not allow features to be easily re-used between different ML pipelines. Features used for training and serving may have different implementations that are not consistent.
In this talk, we will show how ML pipelines can be programmed, end-to-end, in Python. We will show how a Feature Store can provide a natural interface between Data Engineers, who create reusable features from diverse data sources, and Data Scientists, who experiment with predictive models, built from the same features. In an example end-to-end pipeline in Python, we will show how Python dampens the impedance mismatch between Data Engineering and Data Science, enabling Python to become the only language needed for ML pipelines.
Giuseppe Broccolo - Detecting frustration on websites based on digital behaviour
Casper da Costa-Luis - A demo of tqdm