This month we have Jim Savage talking machine learning.

Please note the new location, Civic Hall, courtesy of Bloombergand the later start time of 7:30.

We are just about ready to release tickets to the 2016 New York R Conference! We are going to have a great lineup again this year so stay tuned.

About the talk:

Often, fairly simple parametric models are preferable to complex machine learning methods. Two common examples are in time-series work, where the data are not independent and dependencies need to bemodeledexplicitly, and in causal research, where consistent estimates of the parameters themselves are the primary object of interest. Jim will discuss the usefulness of the Random Forest's proximity matrix in this sort of work. He will present two new methods:

-Proximity matching:A technique for building synthetic control groups for causal analysis on observational data. This method largely overcomes both the dimensionality issues and Smith-Todd critique of other matching methods.

-Analogy weighting:A method for constructing weights to use in time-series models. This weighting scheme allows researchers to give weight to relevant periods of history and ignore irrelevant periods. In doing so, it frees modelers from having to decide on a look-back window for estimation.

About Jim:

Jim Savage is Data Science Lead at Lendable, a frontier markets finance start-up. He recently moved to NY from Australia, where he worked at a leading think-tank on pensions policies, and in macroeconomic modeling at the Australian Treasury. In 2014, he was a fellow at the University of Chicago's Data Science for Social Good fellowship (hurry!2016 applications close soon).

Pizza begins at 7:30, the talk shortly after and then we'll head to a local bar.