[BUG]: pycaret compare_models MAE increase when add installed tpot

[showkeyjar]

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 master branch of pycaret (pip install -U git+https://github.com/pycaret/pycaret.git@master).

Issue Description

I installed pycaret in conda env, and try to compare models with my training data for a regression task,
but it had different results every time, even set session_id and random seed.
It means the best result can not be reproduced.

so, I continue installed tpot to check result, but when I ran pycaret again, the MAE grew 5x than before.

I feel very confused, I'm running pycaret in jupyterlab, and the console has no error.

If I create new env only install pycaret, there is no error any more.

Reproducible Example

`pip install pycaret[full]`

from pycaret.regression import RegressionExperiment

exp.setup(df_samples[ml_sel_cols], target='y', session_id=123)

best = exp.compare_models(exclude=['lr', 'lar', 'llar', 'lasso', 'omp', 'knn', 'huber', 'par', 'ridge', 'dummy', 'catboost'])

`conda install -c conda-forge tpot`

then run upper code again

Expected Behavior

the MAE of two times must be same or near.
and the best result must be reproduced.

Actual Results

but the first best MAE is 0.98

and the second best MAE is 4.99

Installed Versions

python 3.11
conda 22.90
pycaret 3.3.1 and 3.3.0 both tested
tpot 0.12.2
jupyterlab 4.1.5

[showkeyjar]

and I found new bugs, maybe is the same of above: when comparing_models multiple times and testing again, the pipline of the previous version of the model may be used.

it happend in jupyterlab, I used two different datas(data1, data2) to compare_models twice, but the second time predict raise error: no columns XXX found. those missing columns are in data1

it works when I saved model and restart kerenal and rerun it.