Examine Individual Trees in Random Forest - Mahout 0.8

[Tim Peut <ti...@timpeut.com>]

Hi all,

I'm building a random forest in Mahout 0.8 with
org.apache.mahout.classifier.df.mapreduce.BuildForest and need a way to
examine the individual decision trees. For each tree I need to determine
which feature is being used at each node to make the split and what the
decision threshold at each node is.

I tried using the seqdumper tool to examine the output of BuildForest
(forest.seq) but seqdumper gave an error saying that forest.seq is not a
sequence file.

Is it possible to get the information I need about each tree? If so, what's
the simplest way?

Cheers,
Tim

Re: Examine Individual Trees in Random Forest - Mahout 0.8

[Adam Baron <ad...@gmail.com>]

Tim,

Try using org.apache.mahout.classifier.df.tools.ForestVisualizer:
http://shawnwan.wordpress.com/2012/06/01/mahout-0-7-random-forest-examples/

Regards,
           Adam


On Fri, Nov 1, 2013 at 3:33 AM, Tim Peut <ti...@timpeut.com> wrote:

> Hi all,
>
> I'm building a random forest in Mahout 0.8 with
> org.apache.mahout.classifier.df.mapreduce.BuildForest and need a way to
> examine the individual decision trees. For each tree I need to determine
> which feature is being used at each node to make the split and what the
> decision threshold at each node is.
>
> I tried using the seqdumper tool to examine the output of BuildForest
> (forest.seq) but seqdumper gave an error saying that forest.seq is not a
> sequence file.
>
> Is it possible to get the information I need about each tree? If so, what's
> the simplest way?
>
> Cheers,
> Tim
>