List of videos

Extracting insights from a complex machine learning model is not easy hence for many people machine learning models are in a sense black box. This is a problem especially in high stake sectors like banking and healthcare. In this talk we will discuss how we can increase transparency, auditability, and stability of the model using valuable insights we can get from SHAP and explain reasoning behind individual predictions and how this can be aggregated into powerful model-level insights. We will also see the code to calculate SHAP values. Audience level: Intermediate Speaker: Ravi Singh. Data Scientist at HBO Europe developing predictive models and influencing and driving the way the marketing team consume data and insights through highly usable and visual data analysis products.

What is causal inference, and why should data scientists know? × With an explosion of computation power and modern algorithms more and more people are interested in AI, Analytics and Data Science. The Python ecosystem has been one of the most important driver for developing new tools and Python holds the power of modern analytics, but knowing the tools is not enough. Drawing conclusions from data is easy; getting the right conclusions is hard. Causal Inference is the art of drawing robust conclusions from nonexperimental data. This session will be an introduction to the field. Audience level: Novice Speaker: Ludvig Hult

How adjust your AI solution to the new domain? How retrain your model on new industry-specific data? How to increase labeled dataset for minimum cost? Do oracles exist? We will look at these problems from the data protection point of view and bring possible solutions. We will talk about methods, which discuss how to adapt NLP and CV solutions to any type of industry. Speaker: Tetiana Kodliuk

The talk is about sharing the journey & learnings from building a world class AI function that builds scalable software for the entire H&M group. In the talk we will go through the tech stack behind it and some of the key enablers for us to scale and drive value from AI. Speaker: Errol Koolmeister. Errol is the head of AI tech & architecture at H&M group and acting head of Data Science. In his role he is responsible for overseeing and coordinating the AI projects run at H&M. He has spent the last 10+ years in different industries such as banking, telecom and consulting and his main focus has during this time been setting up and scaling AI and big data projects.

Writing highly scalable and provenanceable data pipelines with Kubernetes and Python In this talk we are gonna explore launching and maintaining highly scalable data pipelines using Kubernetes. We are gonna go through the process of setting up a Pachyderm cluster and deploying Python-based data processing workloads. This setup enables teams to develop and maintain very robust data pipelines, with the benefits of autoscaling clusters and quick code iteration. Audience level: Advanced Speaker: Guilherme Caminha, Software Engineer from Brazil. His interests include Scientific / High Performance Computing, Backend Development and Machine Learning.

Python is huge in finance, from banks to hedge-funds to day-traders; it is often the go-to tool to get things done. This talk dives into 'why?'. First looking at several areas where Python is big, what kind of work is involved, and the challenges for developers? Then at what libraries/language aspects make Python particularly suited. Pandas gets particular attention as it is incredibly useful and can make you super productive. We finish up with a demo of how you can use Monte Carlo techniques to build a mini stock market simulation and test out your own trading ideas. After the talk, you should have a better feel for what it's like working with Python in finance and also a good set of starting point if you want to start experimenting. Audience level: Intermediate Speaker: Daniel Roos, with 15+ years of experience building Python/C++ financial systems for global banks and hedge funds. He has recently co-founded Njorda, a retail-focused fintech based in Stockholm.

Generative art leverages computing power to produce elegant artwork, by skillfully controlling the location and magnitude of randomness in color, shape and position. Contemporary artists are more than ever using such principles in their creative workflow, to produce image, video and sound. This talk will give you an overview of the possibilities of using code in an expressive way, and inspire you to start drawing with Python. Audience level: Advanced Speaker: Half Scheidl works as a team lead and project manager in H&M Advanced Analytics, building world-class data platforms and machine learning services with cloud services and software engineering. He is an aspiring digital artist and guest lecturer at Beckmans Designhögskola in Stockholm, engaged in various activities related to digital art. His artwork combines code and technology to create illustrations, animations and interactive installations. Half is the organizer of the Creative Coding Stockholm Meetup group, and has recently exhibited in the Winter Festival at Harpa concert hall in Reykjavik.

Python has become the most widely used programming language. From time to time, employed professionals consider going on their own and become a freelancer. This talk will cover the most demanded Python based services for outsourcing. Anna will also talk about the freelancer mindset together with the legal and economical aspect of being self-employed in Sweden. After the talk, you will be able to see the full picture of freelancing and decide for yourself, if it is for you. Audience level: Novice Speaker: Anna Kazakova Lindegren, Data Scientist, IBM. Anna is a data scientist with a background in accounting and auditing. Before switching to data science in 2017 she had been providing these services to different IT consulting companies and self-employed freelancers for several years. Among her clients were Praqma AB, Tedx Stockholm, A59 and many others.

Stories From Real World Vision Projects How do you implement an application that can pick out the interesting parts of a soccer game? What do you do when you don't have enough data, or when the data is too specific? How do you know that you are actually solving the problem, or even the right problem? Writing a neural network in Python using Keras is very straight forward, you can do it in 20 lines or less, but is that all there is? Welcome to a journey through some real world problems and the thinking when working through them. Speaker: Tess Ferrandez. Tess is a software engineer/data scientist working at Microsoft. She works primarily on computer vision projects and more specifically lately on video action detection projects with some of the largest retailers and media companies in Europe and US. Read her interview here: http://www.pycon.se/blog.html#blog-tess