IRDS: Lab Sessions

The goal of the lab sheets is to give you tutorial-style introductions to a number of toolkits that you may wish to use in your projects and your future career. Each lab sheet will walk you through a different practical data analysis task, in a different software toolbox.

Each sheet will provide pointers to tutorial material / quick reference sheets that provide a list of useful functions and commands. Often, you will find that writing the code would be simple, if you could only figure out what existing functions you need to call. This is normal!

The lab sessions are entirely optional, and they are not marked. You will not hand anything in. Feel free to discuss any issues with your classmates or your tutor.

• Lab 1: MATLAB: Clustering images
• Lab 2: R: Time series data
• Lab 3: Python: Collaborative filtering recipes

Here are a few extra-optional labs that you can go through if they are especially relevant or if you have extra time.

• Lab X1: MALLET and topic modelling

Finally there are some good quality software tools for data science that we don't have time to go through in the lab sessions but are still worth knowing.

• Theano : Neural networks with GPUGPUs. Particularly good for autoencoders.
• Computation Graph Toolkit : Interesting new toolkit based on the design of Theano, but with a faster compiler and better support for mulithreading
• Torch: Machine learning, neural networks, GPUs. I would expect it to be particularly good for ConvNets.
• Caffe -- Fast C++ library, especially well suited for conv nets
• Apache Hadoop: Distributed file system and programming framework. Suited for more coarse grained parallelism than GPGPU.
• Apache Spark: Good for data flows of map-reduce jobs.
• There are a number of new packages for doing probabilistic programming, which is a way of writing programs that programmatically specify graphical models.