[BUG]: Current pycaret-ts-alpha package installs incompatible version of scikit-learn

[rantos9]

pycaret version checks

• I have checked that this issue has not already been reported here.

• I have confirmed this bug exists on the latest version of pycaret.

• I have confirmed this bug exists on the develop branch of pycaret (pip install -U git+https://github.com/pycaret/pycaret.git@develop).

Issue Description

scikit-learn just updated its package to version 1.1.0 and pycaret dosen't seem to be currently compatible with it but pypi package dosen't seem to enforce a version lower than this.

Reproducible Example

from pycaret.time_series import TSForecastingExperiment

Expected Behavior

Importing with no problems

Actual Results

---------------------------------------------------------------------------

ImportError                               Traceback (most recent call last)
Input In [1], in <cell line: 7>()
      5 import numpy as np
      6 import pandas as pd
----> 7 from pycaret.time_series import TSForecastingExperiment
      9 import plotly.express as px
     10 import plotly.graph_objects as go
 
File ~/anaconda3/envs/pycaret-ts/lib/python3.8/site-packages/pycaret/time_series/__init__.py:1, in <module>
----> 1 from pycaret.time_series.forecasting.oop import TSForecastingExperiment
      2 from pycaret.time_series.forecasting.functional import (
      3     setup,
      4     create_model,
   (...)
     25     check_stats,
     26 )
     28 __all__ = [
     29     "TSForecastingExperiment",
     30     "setup",
   (...)
     52     "check_stats",
     53 ]
 
File ~/anaconda3/envs/pycaret-ts/lib/python3.8/site-packages/pycaret/time_series/forecasting/oop.py:16, in <module>
     14 from sklearn.base import clone
     15 from sktime.forecasting.base import ForecastingHorizon
---> 16 from sktime.forecasting.model_selection import (
     17     temporal_train_test_split,
     18     ExpandingWindowSplitter,
     19     SlidingWindowSplitter,
     20 )
     22 from sktime.forecasting.base import BaseForecaster
     24 # from sktime.forecasting.compose import ForecastingPipeline

File ~/anaconda3/envs/pycaret-ts/lib/python3.8/site-packages/sktime/forecasting/model_selection/__init__.py:22, in <module>
     20 from sktime.forecasting.model_selection._split import SlidingWindowSplitter
     21 from sktime.forecasting.model_selection._split import temporal_train_test_split
---> 22 from sktime.forecasting.model_selection._tune import ForecastingGridSearchCV
     23 from sktime.forecasting.model_selection._tune import ForecastingRandomizedSearchCV

File ~/anaconda3/envs/pycaret-ts/lib/python3.8/site-packages/sktime/forecasting/model_selection/_tune.py:13, in <module>
     11 from sklearn.base import clone
     12 from sklearn.model_selection import ParameterGrid, ParameterSampler, check_cv
---> 13 from sklearn.model_selection._search import _check_param_grid
     14 from sklearn.utils.metaestimators import if_delegate_has_method
     16 from sktime.exceptions import NotFittedError

ImportError: cannot import name '_check_param_grid' from 'sklearn.model_selection._search' (/home/ec2-user/anaconda3/envs/pycaret-ts/lib/python3.8/site-packages/sklearn/model_selection/_search.py)

Installed Versions

System:
python: 3.8.13 | packaged by conda-forge | (default, Mar 25 2022, 06:04:18) [GCC 10.3.0]
executable: /home/ec2-user/anaconda3/envs/pycaret-ts/bin/python
machine: Linux-4.14.252-131.483.amzn1.x86_64-x86_64-with-glibc2.10

Python dependencies:
pip: 22.1
setuptools: 60.10.0
pycaret: 3.0.0
sklearn: 1.1.0
sktime: 0.10.1
statsmodels: 0.13.2
numpy: 1.21.6
scipy: 1.7.3
pandas: 1.4.2
matplotlib: 3.5.2
plotly: 5.8.0
joblib: 1.0.1
numba: 0.55.1
mlflow: 1.25.1
lightgbm: 3.3.2
xgboost: 1.6.1
pmdarima: 1.8.5
tbats: Installed but version unavailable
prophet: Not installed
tsfresh: Not installed

[ngupta23]

Thank you for reporting this. This is due to a private method call in one of pycaret's dependencies (sktime): sktime/sktime#2631. Once they fix this, pycaret should be fixed automatically.

For now, can you try limiting sklearn version to < 1.1.0 manually?

[ngupta23]

@Yard1 Since the 3.0.0rc release is still 2-3 weeks out, can we make an updated release for pycaret-ts-alpha from develop but with sklearn pinned to < 1.1.0? This would be a good pretest for 3.0.0rc1 :)

[Yard1]

Sure thing, will do that tomorrow!

[moezali1]

@ngupta23 Is this issue fixed? Should we close it?

[brunobarella]

still with problem! rsrs
@Yard1 my error happened while installing the package.
Do you have any suggestions to fix it momentarily?

[ngupta23]

@brunobarella I do not see any issues anymore. It is working on Google colab. See https://gist.github.com/ngupta23/1dedff5bd94017f5e931614fcdb60187

I will close this now. If the issue persists, please open another issue with all the details to reproduce the error.

Thanks!