plot_tree fails with ValueError Invalid RGBA argument

[giladpn]

Describe the bug

When using plot_tree with filled=True (so the nodes are colored), one sometimes gets a ValueError such as

Invalid RGBA argument: '#cb 3-8d'

The same plot_tree will work fine if filled=False, and draw a decision tree. Below is an example. I attach two files with the dump of the DecisionTree and the list of columns used.

Thank you.

Steps/Code to Reproduce

# BUG: fails with ValueError: Invalid RGBA argument: '#cb 4-8c'
import joblib
from sklearn.tree import plot_tree

dt = joblib.load('_br.dmp')
cols = joblib.load('_cols.dmp')
plot_tree(dt,
          feature_names=cols,
          class_names=['Reject', 'Accept'],
          filled=True, rounded=True,
          impurity=True,
          label='root',
          )

_br.dmp
_cols.dmp

Expected Results

No ValueError should be thrown and the plot should be drawn and correctly colored.

Actual Results

ValueError Invalid RGBA argument: '#cb 3-8d'

Versions

System:
    python: 3.11.4 (main, Jun  7 2023, 12:45:48) [GCC 11.3.0]
executable: /mnt/c/Workspace/py-pip-workspaces/pert.ai.main/.venv/bin/python
   machine: Linux-5.15.153.1-microsoft-standard-WSL2-x86_64-with-glibc2.35

Python dependencies:
      sklearn: 1.5.1
          pip: 24.2
   setuptools: 65.5.0
        numpy: 1.26.4
        scipy: 1.12.0
       Cython: None
       pandas: 2.2.2
   matplotlib: 3.9.1
       joblib: 1.3.2
threadpoolctl: 3.3.0

Built with OpenMP: True

threadpoolctl info:
       user_api: blas
   internal_api: openblas
    num_threads: 8
         prefix: libopenblas
       filepath: /mnt/c/Workspace/py-pip-workspaces/pert.ai.main/.venv/lib/python3.11/site-packages/numpy.libs/libopenblas64_p-r0-0cf96a72.3.23.dev.so
        version: 0.3.23.dev
threading_layer: pthreads
   architecture: SkylakeX

       user_api: blas
   internal_api: openblas
    num_threads: 8
         prefix: libopenblas
       filepath: /mnt/c/Workspace/py-pip-workspaces/pert.ai.main/.venv/lib/python3.11/site-packages/scipy.libs/libopenblasp-r0-23e5df77.3.21.dev.so
        version: 0.3.21.dev
threading_layer: pthreads
   architecture: SkylakeX

       user_api: openmp
   internal_api: openmp
    num_threads: 8
         prefix: libgomp
       filepath: /mnt/c/Workspace/py-pip-workspaces/pert.ai.main/.venv/lib/python3.11/site-packages/scikit_learn.libs/libgomp-a34b3233.so.1.0.0
        version: None

[Siddharth-Latthe-07]

The issue is related to invalid color codes generated for the nodes in the decision tree. This issue typically arises when the generated colors for the nodes are not valid RGBA values.

suggestion and solution:-

1. Check the Decision Tree Classifier
2. Upgrade Scikit-Learn and Matplotlib
3. Generate the Plot Manually
   (Sample snippet:-)

import matplotlib.pyplot as plt
from sklearn.tree import plot_tree

dt = joblib.load('_br.dmp')
cols = joblib.load('_cols.dmp')

fig, ax = plt.subplots(figsize=(12, 12))  # Customize the figure size as needed
plot_tree(dt,
          feature_names=cols,
          class_names=['Reject', 'Accept'],
          filled=True, rounded=True,
          impurity=True,
          label='root',
          ax=ax)
plt.show()

4. Inspect the Generated Colors
5. Custom Color Mapping:-
   (sample code:-)

import matplotlib.pyplot as plt
from sklearn.tree import plot_tree

dt = joblib.load('_br.dmp')
cols = joblib.load('_cols.dmp')

# Define a custom color mapping
custom_colors = ['#FF0000', '#00FF00', '#0000FF', '#FFFF00', '#00FFFF', '#FF00FF']  # Add more colors as needed

fig, ax = plt.subplots(figsize=(12, 12))
plot_tree(dt,
          feature_names=cols,
          class_names=['Reject', 'Accept'],
          filled=True, rounded=True,
          impurity=True,
          label='root',
          ax=ax,
          colors={'color': custom_colors})
plt.show()

Hope this helps, Thanks

[giladpn]

@Siddharth-Latthe-07 thank you for the effort. I tried both of your suggestions and they're both broken, the first has the same exception
'''
ValueError: Invalid RGBA argument: '#cb 3-8d'
'''

while the second suggestion breaks with the exception
'''
TypeError: got an unexpected keyword argument 'colors'
'''

The Decision Tree classifier works, you have its pickle file included in the problem description above.
Matplotlib and scikit-learn versions are indeed the latest, as mentioned above.
I'd appreciate any other ideas. Thanks.

[adam2392]

What is _br.dmp and _cols.dmp? It would help if you can share a MVP with the necessary code using DecisionTreeClassifier in sklearn since loading pickled data arbitrarily is insecure, so one will generally not download and run the code you attached.

[giladpn]

@adam2392 thanks, the _br.dmp is the fitted DecisionTreeClassifier and the _cols.dmp is the array of feature names. This is a specific example, in full detail, that reproduces the issue. I am not sure how else to share it. Is there a different format that is better liked?

For example, do you know if it is possible to persist a fitted DecisionTreeClassifier with JSON? If that works for you, I will do it.

[adam2392]

I meant is there a way to share the relevant code that also fit the DTC? So we can just run it end to end without relying on a pickled file?

[giladpn]

@adam2392 thanks for the effort but sorry - I am stumped. It's part of a large program, with many thousands of lines of code and >100,000 rows of data. Dumping the _br.dmp and _cols.dmp was my way of getting around the complexity.

Is there any other serialization format that would be acceptable?

[adrinjalali]

@giladpn unfortunately we avoid downloading user provided binaries / serialized formats of any kind. Hope you understand.

Therefore a minimal reproducible would be necessary for us to be able to dig deeper and write tests when we find the fix. Closing for now, and happy to reopen if you find a way to provide a minimal reproducer.