19

I am using method on https://machinelearningmastery.com/visualize-gradient-boosting-decision-trees-xgboost-python/ to plot a XGBoost Decision Tree

from numpy import loadtxt
from xgboost import XGBClassifier
from xgboost import plot_tree
import matplotlib.pyplot as plt
# load data
dataset = loadtxt('pima-indians-diabetes.csv', delimiter=",")
# split data into X and y
X = dataset[:,0:8]
y = dataset[:,8]
# fit model no training data
model = XGBClassifier()
model.fit(X, y)
# plot single tree
plot_tree(model)
plt.show()

As I got 150 features,the plot looks quite small for all split points,how to draw a clear one or save in local place or any other ways/ideas could clearly show this ‘tree’ is quite appreciated enter image description here

Leon
  • 389
  • 1
  • 4
  • 14

4 Answers4

21

I had the same problem recently and the only way I found is by trying diffent figure size (it can still be bluery with big figure. For exemple, to plot the 4th tree, use:

fig, ax = plt.subplots(figsize=(30, 30))
xgb.plot_tree(model, num_trees=4, ax=ax)
plt.show()

To save it, you can do

plt.savefig("temp.pdf")

Also, each tree seperates two classes so you have as many tree as class.

Serk
  • 432
  • 4
  • 15
5

To add to Serk's answer, you can also resize the figure before displaying it:

# ...
plot_tree(model)
plt.gcf().set_size_inches(18.5, 10.5)
plt.show()
Learning is a mess
  • 7,479
  • 7
  • 35
  • 71
4

You can try using the to_graphviz method instead - for me it results in a much more clear picture.

xgb.to_graphviz(xg_reg, num_trees=0, rankdir='LR')

However, most likely you will have issues with the size of that output.

In this case follow this: How can I specify the figsize of a graphviz representation of a decision tree?

desertnaut
  • 57,590
  • 26
  • 140
  • 166
morienor
  • 339
  • 1
  • 2
  • 8
  • It's the same thing. [`xgb.plot_tree()` is simply a very thin wrapper to `xgb.to_graphviz()`](https://github.com/dmlc/xgboost/blob/master/python-package/xgboost/plotting.py#L210) – smci Mar 02 '20 at 10:45
  • And it may be slow to show on Jupyter Notebook since svg is hard to handle. – Mehmet Burak Sayıcı Dec 08 '20 at 14:48
3

I found this workaround on github, which also gives better images with the drawback that you have to open the .png file after. 

xgb.plot_tree(bst, num_trees=2)
fig = matplotlib.pyplot.gcf()
fig.set_size_inches(150, 100)
fig.savefig('tree.png')
confused_zebra
  • 116
  • 5