Publications

Previous | item20 | Next

The paper is at the top of
http://www-stat.stanford.edu/~owen/reports

The idea in data squashing is to scale down
a data set to existing algorithms, instead of
scaling up algorithms to large data sets.

Data squashing originated at ATT.

The present paper:

  1)  compares data squashing to other methods
      from sampling, numerical integration, and
      nonparametric likelihood

  2)  tries squashing with boosted trees as
      well as logistic regression

  3)  investigates accuracy of classifications, in
      addition to accuracy of coefficients

-Art Owen

Previous | item20 | Next