3

What do the numbers on the LightGBM plot_tree method represent? As an example, I used the Pima Indian Diabettes dataset and then used the plot_tree method to yield the following: enter image description here

What do the numbers on the leaf nodes represent?

David293836
  • 1,165
  • 2
  • 18
  • 36

2 Answers2

0

Do you mean leaf numbers like "leaf 1", "leaf 2", ect.

They are just labels that represent each leaf output.

To understand it better, I would suggest checking the following URL:

https://machinelearningmastery.com/visualize-gradient-boosting-decision-trees-xgboost-python/

Koorosh Aslansefat
  • 32
  • 2
  • 12
  • No, I am not referring to these leaf numbers. The question is about the the floating values after the leaf numbers. Using the graph above as an example, leaf 0 has a value of -0.668. What does -0.668 mean? – David293836 Feb 04 '21 at 17:35
  • The values should be the "information gain" for each decision. Depending on the settings of the LightGBM, it could also be entropy, Gini impurity, etc. – Koorosh Aslansefat Feb 04 '21 at 20:05
  • Thanks! I'll look it up. – David293836 Feb 08 '21 at 19:51
  • 1
    Please see [here](https://stackoverflow.com/questions/40926340/what-does-the-value-of-leaf-in-the-following-xgboost-model-tree-diagram-means) and [here](https://stats.stackexchange.com/questions/395697/what-is-an-intuitive-interpretation-of-the-leaf-values-in-xgboost-base-learners). Note that those values on the leaves are **not** "information gains" as Koorosh Aslansefat suggested. – Lei Jun 12 '21 at 07:08
0

According to the API description, It is the predicted value.
Actually, It's residual on the leaf.
The each residuals are combined to generate the final estimate.

By the way,
There are many articles on Gradient Boosting Decision Tree Algorithm,
but one of the simplest explanations is here.

masaya
  • 410
  • 2
  • 9
  • 15