[BUG]: Regression Compare Models only returning LightGBM and Dummy

[isaiahxcruz]

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 recently installed the upgraded pycaret via pip3. I created a time series with random values for y and attempted to run a regression experiment. When it runs, it goes through a list of all the models provided but the result is a return of LightGBM and Dummy regressor, nothing else.

Reproducible Example

import pandas as pd
import numpy as np

dates = pd.date_range(start='1/1/2021', end='1/1/2022', freq='D')
vcores = np.random.randint(1, 50, size=len(dates))
data={'dates':dates, 'vcores':vcores}
df = pd.DataFrame(data=data)
df.set_index(dates, inplace=True)
df.drop('dates', axis=1, inplace=True)



train_ind=int(len(df)*0.8)
train=df[:train_ind]
test=df[train_ind:]


from pycaret.regression import *

exp = setup(train, train_size=0.8, target='vcores', data_split_shuffle=False, fold_strategy='timeseries')
compare_models()

Expected Behavior

huber	Huber Regressor	0.0494	0.0130	0.0713	0.9268	0.0147	0.0127	0.0290
Lasso Regression	0.0653	0.0335	0.1148	0.8545	0.0255	0.0196	0.0070
Elastic Net	0.0652	0.0335	0.1148	0.8544	0.0255	0.0196	0.0070
Gradient Boosting Regressor	0.2910	0.3991	0.3570	-0.4301	0.0685	0.0863	0.1210
Extra Trees Regressor	0.3423	0.5543	0.4055	-1.0313	0.0777	0.1036	0.5070
Orthogonal Matching Pursuit	0.1437	0.5384	0.3034	-1.0484	0.0181	0.0469	0.0060
AdaBoost Regressor	0.3493	0.5595	0.4185	-1.0499	0.0792	0.1040	0.0890
Random Forest Regressor	0.3658	0.5649	0.4228	-1.0551	0.0815	0.1099	0.4960
Decision Tree Regressor	0.3968	0.6173	0.4948	-1.4153	0.0901	0.1118	0.0080
Extreme Gradient Boosting	0.4873	0.8058	0.5700	-2.3462	0.1048	0.1401	0.3240
CatBoost Regressor	0.5834	0.9216	0.6547	-2.4012	0.1247	0.1731	1.9830
Light Gradient Boosting Machine	0.5174	0.9572	0.5895	-2.4582	0.1107	0.1528	0.0230
Passive Aggressive Regressor	0.9121	1.5765	0.9701	-5.6274	0.1817	0.2434	0.0080
K Neighbors Regressor	0.7623	1.1585	0.8794	-7.0430	0.1613	0.2138	0.3310
Lasso Least Angle Regression	1.3113	2.9353	1.4034	-11.8521	0.2544	0.3879	0.0050
Linear Regression	4.0729	145.1286	5.7263	-483.7128	0.3858	0.9488	0.0060
Bayesian Ridge	7.3757	2826.3201	17.6026	-10669.0481	0.3243	2.7772	0.0320
Ridge Regression	8.3168	4331.3753	20.8806	-16361.0021	0.2116	3.1673	0.0050
Least Angle Regression	18583978932258.3008	5903475984839208765730324480.0000	24454644538648.7266	-8865067586334214311708196864.0000	10.0498	5121901449923.7939	0.0160

Actual Results

lightgbm	Light Gradient Boosting Machine	12.2235	203.2234	14.1625	-0.0368	0.7676	1.4719	0.0140
dummy	Dummy Regressor	12.2235	203.2234	14.1625	-0.0368	0.7676	1.4719	0.0060

Installed Versions

2.3.10

[joncassidy]

I have the same issue when adding many (100+) float16 features, but the issue does not happen if the features are float instead.

[celestinoxp]

Hi,
are you using pycaret 2.3.10? i think it's problem from scikit-learn, because pycaret uses "old" 0.24.2 version of scikit-learn.
Scikit-learn have a lot of bugfixes since 0.24.2. I think i have seen something like there in scikit bugfixes...

Suggestion: install latest pycaret (rc3 at this time) and it will install latest scikit-learn. Test if works. After that tell me if works.

Note: To avoid problems with packages uninstall first pycaret and sckit-learn, then install latest pycaret (it will install scikit-learn automatically)

if you don't know how to do it, use these commands:

pip uninstall pycaret -y
pip uninstall scikit-learn -y
pip install --pre pycaret or pip install --pre pycaret[full]

Important: if you don't use python 3.8 or python 3.9 I recommend creating a new environment with one of these versions.

[moezali1]

By changing few things in setup I was able to run your code successfully using 3.0rc4

pip install --pre pycaret

Check out the Notebook.