I got a second look at VNI’s product this week – I took a first look last year. VNI has been continuing to OEM its products to folks from around the world. Their customers are in many different areas like Investor Analytics (SaaS for financial risk management optimization), Moore Nanotechnology Systems machinery and RiskMetrics Group (who added a stress testing capability for portfolios). Some of their OEMs expose the functionality (analytics and math routines) pretty directly to users while others are more embedded/hidden. An increasing number of their customers are SaaS providers and the trend is towards OEMs who embed the functionality more deeply with less exposed. They offer a mix of algorithms for math and for data mining with many routines getting reused such as optimization routines, adjustments to time series etc. They see an increasing demand for near real-time performance (in time series forecasting for example) and this is driving new algorithm usage and demand for performance and parallelism. For instance, overnight batch processes for portfolio optimization moving to minutes for intra-day strategies.
I got a couple of updates. First they talked about V 7.0 of the IMSL…
I got a second look at VNI’s product this week – I took a first look last year. VNI has been continuing to OEM its products to folks from around the world. Their customers are in many different areas like Investor Analytics (SaaS for financial risk management optimization), Moore Nanotechnology Systems machinery and RiskMetrics Group (who added a stress testing capability for portfolios). Some of their OEMs expose the functionality (analytics and math routines) pretty directly to users while others are more embedded/hidden. An increasing number of their customers are SaaS providers and the trend is towards OEMs who embed the functionality more deeply with less exposed. They offer a mix of algorithms for math and for data mining with many routines getting reused such as optimization routines, adjustments to time series etc. They see an increasing demand for near real-time performance (in time series forecasting for example) and this is driving new algorithm usage and demand for performance and parallelism. For instance, overnight batch processes for portfolio optimization moving to minutes for intra-day strategies.
More Read
I got a couple of updates. First they talked about V 7.0 of the IMSL C Numerical Library – a big release launched in mid-November. This had:
- Parallelization for multi-core architectures. Good performance improvements across the board. Moves some problems from the server to the desktop – makes it easier to solve some problems. Many OEMs are running as a desktop app so this is critical for their clients.
- New algorithms – genetic algorithms, naïve bayes, Feyman-Kac (solver for Black-Scholes among others but also partial differential equation solver).
- General updates to existing algorithms
The second update was PyIMSL Studio released February 10. This is designed to turn prototypes into production – helping modelers stay focused on the best model while allowing developers to choose the best implementation. Python seems to be popular among analytic modelers in my experience and VNI said that 2/3 of existing customers had some use of python already. Their insight was that developing a prototyping environment was easier than developing the libraries for deployment so they developed a prototyping environment for their existing IMSL C Library. Making a prototyping tool that allows implementation is key because C/C++ skills are not as common in these high end analytic problems making it hard to find developers for these kinds of systems. The product has three components:
- PyIMSL – C libraries with Python wrappers
- IMSL Data Utilities- C libraries for handling data
- Python studio that packages up a bunch of open source Python components including Eclipse and IPython IDE
Once the modeler is done can hand it over to the developer who can use Eclipse to redo the glue but all the calls are the same thanks to the shared library.
It was good to hear from VNI again and good to see them offering a tool to help close the modeling to deployment gap. This is a big focus in the analytics space, as it should be, and I thought their support for Python as a prototyping/modeling environment was particularly interesting.
