mikeczech / pyvrank

Python package for predicting rankings of software verification competitions.

Hello! I'm a machine learning engineer and data scientist from Hamburg. Diving deep into data, crafting machine learning models, and bringing real-world applications to life is what I love. Always eager to learn and share!

Short CV

• Machine Learning Engineer / Data Scientist at New Work SE (Oct 2021 - Present).
• (Senior) Machine Learning Engineer at Otto Group data.works GmbH (Jun 2018 - Sep 2021).
• Consultant Developer at ThoughtWorks Germany GmbH (Aug 2016 - May 2018).
• Research and tutoring roles at Paderborn University (Apr 2011 - Sep 2015).
• Software Developer at Busitec GmbH (Aug 2007 - Oct 2009).

Open Source Contributions

• Clash: A library for running jobs on the Google Compute Engine.
• Initial author of bquest: A testing framework for BigQuery.

Interests

• Python, SQL, C++
• Machine Learning
• Software Development Practices
• Cloud Platforms
• Mentoring
• Climbing / Bouldering

pyvrank's Issues

IndexError in Weisfeiler Lehman graph kernel

Traceback (most recent call last): File "/Users/cedricrichter/Documents/Arbeit/Ranking/PyPRSVT/scripts/model_induction.py", line 48, in dump_gram K = kernel.compare_list_normalized(graphs, types, h, D) File "/Users/cedricrichter/Documents/Arbeit/Ranking/PyPRSVT/PyPRSVT/gk/GK_WL.py", line 51, in compare_list_normalized k = self.compare_list(graph_list, types, h, D) File "/Users/cedricrichter/Documents/Arbeit/Ranking/PyPRSVT/PyPRSVT/gk/GK_WL.py", line 109, in compare_list phi[self._compress(long_label), i] += 1 IndexError: index 235 is out of bounds for axis 0 with size 234

This error occurs when computing the normalized gram matrix for h = 2 and D = 0.
For the graph computation the xml file in the static folder was used.

The cause may be:

• all_graphs_number_of_nodes is not an upper bound (unlikely),
• for some reason all_graphs_number_of_nodes is not correctly computed,
• in __label_lookup__ are entries which don't belong to the map,
• (unreasonably) numpy.zeros(...) changes its behaviour in a newer version.

graphs.zip contains all files used for the gram matrix computation.