MapReduce requires the data chunks to be processed independently. This processing model is unsuitable for many graph models in which a calculation often requires knowledge about calculations done for other nodes. This issue is exactly what Pregel tackles with message passing.

Configuration files can get messy when dealing with a large number of external parameters. The built-in solutions like argparse might not be scalable or clean enough to manage external parameters. In this post I look at an alternative YAML based solution that implicitly passes configuration to functions.

A very quick introduction to Keras and common architecture types with short examples. The slides include code snippets and tips on training neural networks.

I cover useful information from running batch jobs to introduction for TensorFlow and Keras. These are aimed at students who have machine learning based projects.

Virtualenv provides a very good way to isolate packages. But sometimes it is better to install packages shared across projects in a custom location.