Is there an app for a decision tree graph?

[ericchaucl]

Hi,

After building a machine learning model using DecisionTreeRegressor or DecisionTreeClassifier, I can use the "| summary" command to list out the content.

Is there any better way to visualize the tree instead of row by row? Say, graphviz can plot a tree nicely (see picture as reference). Any customized codes or Splunk app can do this?

[oda · Accepted solution]

https://splunkbase.splunk.com/app/3379/
I think that you can do it with this APPS!

[ericchaucl]

Thanks for your quick reply! The app looks pretty nice and will have a try on it. Seems I need to write codes to transform the results to fit into this chart.

Please forgive my laziness. Just wonder if there is any existing app/code that already can present the summary results from Machine Learning Toolkit directly? As the output from Splunk is in a standard format like below, hopefully Splunk or some developer has already write the code to present it in a nicer way:

Decision Tree Summary
|-- count:10 split:age<=22.500 value:0.404 impurity:0.036
|------ count:1 split:N/A - Leaf node value:0.100 impurity:0.000
|------ count:9 split:age<=46.500 value:0.438 impurity:0.028
|---------- count:8 split:age<=41.500 value:0.475 impurity:0.019
|-------------- count:7 split:age<=40.500 value:0.429 impurity:0.005
|------------------ count:6 split:sex<=1.500 value:0.413 impurity:0.004
|---------------------- count:3 split:age<=35.000 value:0.437 impurity:0.005
|---------------------- count:3 split:age<=36.000 value:0.389 impurity:0.001
|------------------ count:1 split:N/A - Leaf node value:0.524 impurity:0.000
|-------------- count:1 split:N/A - Leaf node value:0.800 impurity:-0.000
|---------- count:1 split:N/A - Leaf node value:0.136 impurity:0.000

[oda]

I'm sorry.
I could not understand what you wanted.
What can I cooperate with you?

[ericchaucl]

The "Decision Tree Summary" example shown in my post is from Splunk directly, using the command "| summary "

It is presented like a tree in text format, using "|----". Can simplify an example like this:

Decision Tree Summary
|-- Node A
|------ Leaf Node B (from A)
|------ Node C (from A)
|---------- Leaf Node D (from C)
|---------- Leaf Node E (from C)

Just wonder if there is there is any existing work done to convert these records into a more presentable graph.

[oda]

I understand!
Thank you!
Unfortunately we can not find such a visualization at present.
There seems to be only processing by oneself.

[ericchaucl]

Well noted, thanks! Hopefully Splunk will consider this in future release of their Machine Learning Toolkit.

[oda]

I also hope that visualization will increase a little more.
Can you accept or upvote if you do not mind?

[ericchaucl]

Would like to wait a few more days and see if there is additional suggestions from others. Let me upvote your answer first.

[oda]

Thank you!
I hope there is a better proposal!