[Research] Auton Lab meeting next week

[Artur Dubrawski]

Dear Autonians,

Now with the fall foliage past its peak, everyone in Pittsburgh
can see more structural detail of trees growing around.
It feels appropriate and timely for computer scientists to relate 
to this phenomenon. Fundamental questions, such as why the "real" trees 
have roots in the bottom and leaves up on top, entirely opposite to how
they are in computers, as well as other crucial and relevant issues, 
will be addressed by the leading Auton Lab tree grower Dr. Daria Sorokina.

When:     Wednesday November 4th at noon
Where:    NSH 1507
Food:     yes
Abstract: see below

See you all there!
Artur

----
TreeExtra package: prediction and descriptive analysis with ensembles of
trees.

In this talk I am going to describe a C++ package I have been developing
over the past two years. TreeExtra contains a number of general data mining
tools relevant for a large range of problems.

- Additive Groves
	 A powerful ensemble of trees, significantly outperforms other
ensembles on regression and is highly competitive on classification
problems. TreeExtra implementation provides several modes of training,
automatic parameter choice and parallelization tools.

- Bagged trees with fast feature evaluation. 
	Bagged trees is a fast and simple ensemble, in our implementation it
is equipped with built-in feature importance ranking.

- Backward elimination feature selection
	Thorough feature selection, leaves only a small set of truly
important non-redundant features.

- Interaction detection
	Detects which features are involved in complex effects and need to
be studied together

- Effect plots (or partial dependence plots)
	Visualization of an effect of a single variable in a given model

- Interaction plots
	Visualization of a joint effect of two variables in a model along
with the joint distribution of those variables