Husband
Father
Machine learning and probabilistic modeling scientist at Charles River Analytics and research fellow at the Vermont Complex Systems Center
Former lead data scientist at MassMutual and trombonist

From time to time I will write about things. I will list the things here.

Keep using plate notation Someone was wrong on the internet. In particular, someone expressed their (questionable) beliefs about plate notation. These are my feelings about that.
Theoretical data science background I: functional optimization First post in a (maybe) multipost series of theoretical background for data scince. Here lie Euler-Lagrange equations, the general supervised learning problem, and time series smoothing.
Perturbing correlation matrices: on what we can do when we are presented with a correlation matrix and not with the data that generated it
Continuum rich-get-richer dynamics and survival theory: on the non-obvious connection between the kernel function of a rich-get-richer process and the hazard function of a random variable.

stsb2 Structural time series, round 2. Implements a grammar over structural time series models and an associated modeling + inference library.
verdantcurve: agent-based market model created in collaboration with Colin Van Oort. This package contains
- A heterogeneous and extensible set of algorithmic trading agents
- A matching engine that implements a frequent batch auction, a new type of auction mechanism that greatly reduces the effectiveness of high-frequency trading strategies
- A orderbook and multiple types of orders
- A market construct to facilitate interaction between agents and the matching engine
discrete-shocklet-transform: qualitative, shape-based, timescale-independent time series similarity search algorithm