Error converting sklearn linear regression model

[IAINATTWATER]

❓Question

System Information

• If applicable

Conversion fails with - Exception has occurred: NameError name '_tree' is not defined.

Model is a linear regression but converter appears to be looking for a decision tree one. There's also a warning about scikit-learn version 0.22.1 is not supported. Minimum required version: 0.17. Maximum required version: 0.19.2. Disabling scikit-learn conversion API.

anyone seen this?

Iain

[anilkatti]

@IAINATTWATER thanks for reporting. We are looking into this.

[sriharsha-sammeta]

Is there a resolution for this ?

[IAINATTWATER]

Not that I’m aware of.

Iain.

On Oct 26, 2020, at 6:51 PM, Harsha Sammeta [email protected] wrote:

 Is there a resolution for this ?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[steven-dicristofaro]

+Also running into this issue when converting a RandomForestClassifier created in sklearn 0.23.2 (python 3.7). Coremltools conversion is being disabled due to the versioning mismatch.

Downgraded sklearn to 0.19.2 at the moment as a workaround, but this issue needs to be addressed as soon possible.

[imcauley]

Were you able to downgrade sklearn via pip? Whenever I try to downgrade to 0.19.2 or lower it fails.

[steven-dicristofaro]

I use a PyCharm managed venv, but I just spun up another one through the terminal and ran:

pip install 'scikit-learn==0.19.2' 

and it worked fine. You could also try using the --force-reinstall flag if that would help.

[IAINATTWATER]

Thanks. Will give it a try.

Sent from my iPhone

On Nov 10, 2020, at 6:46 PM, steven-dicristofaro [email protected] wrote:

 I use a PyCharm managed venv, but I just spun up another one through the terminal and ran:

pip install 'scikit-learn==0.19.2' and it worked fine. You could also try using the --force-reinstall flag if that would help.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[steven-dicristofaro]

also for those who have some flexibility in their model deployment form factor - just discovered this library for python that looks mighty interesting https://github.com/BayesWitnesses/m2cgen#supported-models . Haven't tried it out yet, but IN THEORY it seems we could just transpile to C and bring that into a swift app with a bridging header or via .podspec.

+Java transpiling for any Android guys out there that need this to be X-platform

[TobyRoseman]

At least for for linear regression, it seems to work fine with the most recent version of Scikit Learn (0.24.1) if you just update the max supported version.

[Eddy872]

Hi and sorry for the post but I have the same problem for SVC model from sklearn models. I load my model with pickle because it's also pre-trained, the model works on conda. But when I want to convert this to ml, I have : name '_tree' is not defined Anyone has the same problem with SVC ? Sorry again for the post :)

[imcauley]

@Eddy872 This was my exact problem. I wish I had a better solution for you but I ended up just switching to a model that was actually supported. It's been so long I can't quite remember which one I used. Sorry to be the bearer of bad news :(

[Eddy872]

@imcauley Thank you, maybe you can help me, which models are supported ? models from sklearn ?

[scaiciia]

I resolved! The problem is the Python version. You need creating a PyCharm Project with python version 3.7 and install scikit-learn 0.19.2

[TobyRoseman]

We now support the latest version of Scikit-Learn (1.1.2). So I believe this is fixed.

Can anyone confirm this is fixed? Or give us step to reproduce the issue, if it is still an issue.

[TobyRoseman]

Since I have not heard anything here, I'm going to close this issue. If this has not been fixed by supporting a new version of scikit-learn, please let us know.

[rromanchuk]

scikit-learn version 1.3.0 is not supported. Minimum required version: 0.17. Maximum required version: 1.1.2. Disabling scikit-learn conversion API.

Same thing again, and probably will keep being a thing since it's a hardcoded semver

coremltools/_deps/__init__.py line 58 if anyone needs to get around the pin

[citruslab]

Getting error when using flag --quantize-nbits 6. ValueError: Input contains NaN, infinity or a value too large for dtype('float64'). I tried downgrading python, torch and scikit-learn but nothing changes. Anyone has a tip on how to solve this?

[anz-ryu]

I got

  File "/Users/user2/Documents/anchorz/data_research/lib/action_detection/batch_convert/training.py", line 67, in use_sklearn
    coreml_model = ct.converters.sklearn.convert(gb_clf)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/sklearn/_converter.py", line 148, in convert
    from ._converter_internal import _convert_sklearn_model
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/sklearn/_converter_internal.py", line 24, in <module>
    from . import (_SVC, _SVR, _decision_tree_classifier, _decision_tree_regressor,
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/sklearn/_decision_tree_regressor.py", line 16, in <module>
    sklearn_class = _tree.DecisionTreeRegressor
                    ^^^^^
NameError: name '_tree' is not defined

using: scikit-learn==1.3.0 python==3.11.5

---

resolved it by downgrade to scikit-learn==1.1.2 and python==3.9.6.