Welcome to Umlaut

What is Umlaut?

Umlaut is an end-to-end machine learning benchmarking tool for Python, designed to be applicable in every step of a machine learning pipeline. It includes a Python package that contains functions and classes for measuring metrics of the system and model. The collected measurement values can be visualised and compared using a CLI included in the package.

What can be measured using Umlaut?

Using Umlaut, you can collect and visualise the following metrics:

  • CPU usage

  • Memory consumption

  • Power consumption

  • Runtime

  • Throughput

  • Latency

  • Time-To-Accuracy

  • Loss

  • Confusion Matrix

How do I use it?

We have prepared an example that will guide you through the usage of our tool. We also recommend you have a look at the module reference.

What’s with the name?

Umlaut stands for Universal Machine Learning Analysis Utility.

Who developed it?

Umlaut was developed by Christian Jacob, Jost Morgenstern, Willi Rieck and Jonas Schulze as part of their bachelor project at HPIDES.