In this project I demonstrate how to deploy a ML model as a microservice. The project contains two parts: a simple frontend built with Streamlit and an API built with the FastAPI framework. The frontend is used to acquire information from the user on characteristics that are used as input for the predictive model. The API consists of two endpoints - one for preprocessing the data and one for making predictions. The frontend sends requests to this API and gets the preprocessed data and the predictions as responses. Moreover, I show how to containerize both parts using Docker, and eventually how to deploy them together as a multi-container application with Docker-Compose.

One important task for data scientists is visualizing data in a form that it can be accessed and used by stakeholders. A common way to achieve this, is by using dashboards that display relevant data clearly and in a compact form. In this blogpost I build a dashboard on the current Covid-19 situation in Germany. The dashboard provides information on the most relevant figures, is interactive, and is automatically updated with the newest data on a daily basis using GitHub Actions.

In the course of data science projects, often a large number of different combinations of algorithms, preprocessing steps, and featuers is evaluated to find the best performing model. In this blogpost I introduce a simple and automated way of keeping track of different models’ performances using the MLflow framework.

This blogpost provides an introduction to and brief explanation of a recently developed method combining Machine Learning algorithms and classical ideas from econometrics to obtain statistically valid inferential statements on heterogeneous treatment effects in Randomized Controlled Trials (RCTs).

This blogpost contains the code for a Web Scraping project to extract the soccer metrics expected goals (xG) and expected goals against (xGA) from understat.com as well as additional match statistics. The code works for all listed leagues and seasons.

In this blogpost I describe the process of building a good-looking ML App using Streamlit. The app again implements the genre classification model, however the whole process is much easier and faster.

After building and training a ML model, the next step to actually generate business value is to deploy the model so that it can be used in production. In this blog I build a webapp and deploy the genre classification model built in the last blogpost on Heroku.

In this project I build a neural net to classify songs into one of three genres based on their lyrics. The lyrics are obtained via the lyricsgenius package and the model is built with the powerful fastai library.

The aim of the project is to predict disease rates, more precisely influenza-like illnesses rates (ILI rates). On the one hand historical time series data were used, on the other hand the data basis was extended by Google Trends data.

This blogpost contains code for the Kaggle Titanic challenge. It is a well-known beginner challenge and its goal is to build a binary classification model to predict if passengers would have survived the Titanic disaster or not. The focus of this blogpost is on Exploratory Data Analysis and some Feature Engineering. Later on, I built and deployed a small ML app using Streamlit and Streamlit Sharing.