okay thanks for verifying , I will recheck from end and update.

from scikit-learn-intelex.

@darshanp4 please use verbose mode for getting more informative logs why you are falling back to stock sklearn.

from scikit-learn-intelex.

@samir-nasibli only seeing following logs on console :

Intel(R) Extension for Scikit-learn* enabled (https://github.com/intel/scikit-learn-intelex)
INFO:sklearnex: sklearn.utils.validation._assert_all_finite: running accelerated version on CPU

from scikit-learn-intelex.

is verbose supported for SVR ?
seen this issue : #1012

from scikit-learn-intelex.

@darshanp4 thank you for the report. I will investigate and share.

from scikit-learn-intelex.

@darshanp4, SVC/SVR with gamma='auto' parameter is generally supported in 2024.0.1 version, so I guess the reason is on input data side. Can you share what data type is used for x and y?

from scikit-learn-intelex.

data_729para.csv
attached data file
data = pd.read_csv(f"data_729para.csv")
x = data.iloc[:, data.columns!='H'].to_numpy()
y = data['H'].to_numpy()

from scikit-learn-intelex.

data_729para.csv attached data file data = pd.read_csv(f"data_729para.csv") x = data.iloc[:, data.columns!='H'].to_numpy() y = data['H'].to_numpy()

@darshanp4 I have checked it with scikit-learn-intelex 2024.0.1 and daal4py 2024.0.1. But can not reproduce your issue. It works fine on my env:

daal4py                   2024.0.1         py310_intel_25    intel
numpy                     1.22.4                   pypi_0    pypi
pandas                    2.1.3                    pypi_0    pypi
python                    3.10.13              h955ad1f_0  
scikit-learn              1.3.2                    pypi_0    pypi
scikit-learn-intelex      2024.0.1         py310_intel_25    intel

I was using this script for validation:

import logging

import numpy as np
import pandas as pd


logging.getLogger("sklearnex").setLevel(logging.INFO)

data = pd.read_csv("../data.csv")

from sklearnex import patch_sklearn

patch_sklearn()

X = data.iloc[:, data.columns!='H'].to_numpy()
y = data['H'].to_numpy()
print("x data type: ", X.dtype)
print("y data type: ", y.dtype)

from sklearn.svm import SVR
sv_params = {"gamma": 'auto'}
model = SVR(**sv_params)
model.fit(X, y)

Output:

Intel(R) Extension for Scikit-learn* enabled (https://github.com/intel/scikit-learn-intelex)
x data type:  float64                                      
y data type:  int64                                              
INFO:sklearnex: sklearn.svm.SVR.fit: running accelerated version on CPU

So actually, it doesn't fallback to stock sklearn.

from scikit-learn-intelex.

That verbose output I also see on console, but did you check that it actually using accelerated oneDAL. If you profile it uses the stock. That's what I reported initial. Can you paste the perf report as well of your run? To confirm it uses the intelex version.

from scikit-learn-intelex.

That verbose output I also see on console, but did you check that it actually using accelerated oneDAL. If you profile it uses the stock. That's what I reported initial. Can you paste the perf report as well of your run? To confirm it uses the intelex version.

Please be careful, your report here doesn't indicate running it on onedal accelerated SVR.fit:

INFO:sklearnex: sklearn.utils.validation._assert_all_finite: running accelerated version on CPU

In contrast, my logs show this:

INFO:sklearnex: sklearn.svm.SVR.fit: running accelerated version on CPU

Also profiler shows that it uses onedal branch:

        1    0.000    0.000    0.000    0.000 _common.py:60(_onedal_cpu_supported)
        1    0.000    0.000    0.000    0.000 svm.py:217(_get_onedal_params)
        1    0.000    0.000    0.146    0.146 svr.py:145(_onedal_fit)
        4    0.000    0.000    0.000    0.000 {built-in method onedal._onedal_py_host.from_table}
        1    0.145    0.145    0.145    0.145 {built-in method onedal._onedal_py_host.svm.regression.train}
        3    0.000    0.000    0.000    0.000 {built-in method onedal._onedal_py_host.to_table}

from scikit-learn-intelex.

Sorry for late update , issue was due to using below flag

os.environ['NO_DIST']='1'

os.environ['NO_STREAM']='1'

os.environ['OFF_ONEDAL_IFACE']='1'

from scikit-learn-intelex.