read-me.txt

来自「垃圾邮件的近几年的研究趋势 及方向 以及改进方法」· 文本 代码 · 共 23 行

These files are pre-processed by joining the participants' entries
with the test set and preparing the output to be used by the 'perf'
program <http://kodiak.cs.cornell.edu/kddcup/software.html>.

# How to obtain the area under the ROC curve
(for i in *.perf; do echo -n "$i "; perf -ROC < $i; done) | awk '{print $1, sprintf("%.3f", $3); }' | sort -rnk2

geng.txt.perf 0.848
tang.txt.perf 0.824
abernethy.txt.perf 0.809
siklosi2.txt.perf 0.796
bauman.txt.perf 0.783
skvortsov.txt.perf 0.731
siklosi1.txt.perf 0.726

Note: apparently the AUC of the first system was 0.848 -> 0.85 when
approximated with two digits (in the slides presented at the workshop
it read 0.84). This obviously does not affect the ordering of the results.

--
Thu May  8 11:30:05 CLT 2008