It may have been the first day of spring but it was a cold, wet night for the September Melbourne Python Users Group meeting. We had at least twenty five people attend for three talks. Thanks to Inspire 9 for hosting and Planet Innovation for sponsoring the meeting.
1. Javier Candeira – What’s new in Python
Javier updated us on what’s new and interesting in Python over the last month. His slides are online.
2. Andy Kitchen – Pattern Recognition: machine learning on GPUs in the cloud
Andy is a developer at Silverpond and walked us through using Theano for deep learning applications.
- Deep learning is (very) computationally intensive
- Hyperdimensional optimisation, mostly dense-matrix multiplication – use GPUs because they are often 40x faster than CPU
- GPU programming tricky but getting easier (eg. CUDA and Theano)
- Amazon EC2 for about $5/hr: NVIDIA AMI with CUDA, surprisingly smooth!
- Theano: write once for CPU or GPU (via CUDA backend) by building expression trees
- Demo: 14.23 secs with CPU, 1.14 secs with GPU
- (Sidenote: it is also worth looking at NumbaPro from Continuum)
3. Jason King – BDD using Behave with Selenium
Jason took us through some recent work he has been doing at Common Code for testing an online medical appointment booking system. He has been using Selenium and Behave, a BDD framework that uses decorators to determine what test to execute.