tpot
tpot copied to clipboard
EpistasisLab
Import issue with tpot 0.12.0 and sklearn 1.0.2
[provide general introduction to the issue and why it is relevant to this repository]
Context of the issue
tpot 0.12.0 gives error with scikit-learn v1.0.2, but loads ok with scikit-learn v1.1.3
Process to reproduce the issue
import tpot
Expected result
TPOT loads without error
Current result
╭─────────────────────────────── Traceback (most recent call last) ────────────────────────────────╮
│ in <module>:1 │
│ │
│ C:\...\lib\site-packages\tpot\__init__.py:27 in │
│ <module> │
│ │
│ 24 """ │
│ 25 │
│ 26 from ._version import __version__ │
│ ❱ 27 from .tpot import TPOTClassifier, TPOTRegressor │
│ 28 from .driver import main │
│ 29 │
│ │
│ C:\...\lib\site-packages\tpot\tpot.py:31 in <module> │
│ │
│ 28 │
│ 29 import numpy as np │
│ 30 │
│ ❱ 31 from .base import TPOTBase │
│ 32 from .config.classifier import classifier_config_dict │
│ 33 from .config.regressor import regressor_config_dict │
│ 34 │
│ │
│ C:\...\lib\site-packages\tpot\base.py:83 in <module> │
│ │
│ 80 from .config.classifier_cuml import classifier_config_cuml │
│ 81 from .config.regressor_cuml import regressor_config_cuml │
│ 82 │
│ ❱ 83 from .metrics import SCORERS │
│ 84 from .gp_types import Output_Array │
│ 85 from .gp_deap import ( │
│ 86 │ eaMuPlusLambda, │
│ │
│ C:\...\lib\site-packages\tpot\metrics.py:27 in │
│ <module> │
│ │
│ 24 """ │
│ 25 │
│ 26 import numpy as np │
│ ❱ 27 from sklearn.metrics import get_scorer, get_scorer_names, make_scorer │
│ 28 │
│ 29 │
│ 30 def balanced_accuracy(y_true, y_pred): │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
ImportError: cannot import name 'get_scorer_names' from 'sklearn.metrics'
Possible fix
Either increase the required version of sckit-learn or fix tpot to keep current dependency.
name of issue screenshot
[if relevant, include a screenshot]
Traceback (most recent call last):
File ".\forward_predict.py", line 16, in <module>
from tpot.builtins import StackingEstimator
File "C:\Users\chalu\AppData\Local\Programs\Python\Python38\lib\site-packages\tpot\__init__.py", line 27, in <module>
from .tpot import TPOTClassifier, TPOTRegressor
File "C:\Users\chalu\AppData\Local\Programs\Python\Python38\lib\site-packages\tpot\tpot.py", line 31, in <module>
from .base import TPOTBase
File "C:\Users\chalu\AppData\Local\Programs\Python\Python38\lib\site-packages\tpot\base.py", line 82, in <module>
from .metrics import SCORERS
File "C:\Users\chalu\AppData\Local\Programs\Python\Python38\lib\site-packages\tpot\metrics.py", line 27, in <module>
from sklearn.metrics import make_scorer, SCORERS
ImportError: cannot import name 'SCORERS' from 'sklearn.metrics' (C:\Users\chalu\AppData\Local\Programs\Python\Python38\lib\site-packages\sklearn\metrics\__init__.py)
TPOT doesn't like latest version of scikit-learn, version 1.3.0 I believe. SCORERS isn't a part of the sklearn.metrics class anymore. Have to revert to older versions of scikit-learn.
This is actually not the latest version of scikit-learn, but an older version.
Running pip install scikit-learn installs version 1.3.0 as of 2 days ago, so that seems to be the most recent stable version they're using. I see there's dev/beta versions higher than that, but I would assume that most people who install it will get version 1.3.0 by running pip install. Either way, the issue persists.
Add te methodget_scorer_names()in _scorer.pyinside sklearn directory and it should work if you try to import again the TPOTClassifier:
def get_scorer_names():
"""Get the names of all available scorers.
These names can be passed to :func:`~sklearn.metrics.get_scorer` to
retrieve the scorer object.
Returns
-------
list of str
Names of all available scorers.
"""
return sorted(SCORERS.keys())
