About model persistence

[gaosiy]

Hi everyone, I use intelex to speed up my svm training. But when I try to save the model and use it on another device, it report a error "ModuleNotFoundError: No module named 'sklearnex' ". I think the intelex may add something to the model which I do not need. because the new device is an arm-linux device, I can not install sklearnex on it.

My demo code show here, first, I use svm without intelex and save the model.
`from sklearn.svm import SVC
from sklearn import metrics
from sklearn.datasets import fetch_openml
from sklearn.model_selection import train_test_split

x, y = fetch_openml(name='a9a', return_X_y=True)
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=42)
params = {
'C': 100.0,
'kernel': 'rbf',
'gamma': 'scale'
}

classifier = SVC(**params).fit(x_train, y_train)

from joblib import dump, load
dump(classifier, './model/svm_normal.joblib') `

Then I used intelex and saved the model
from sklearnex import patch_sklearn patch_sklearn() params = { 'C': 100.0, 'kernel': 'rbf', 'gamma': 'scale' } start = timer() classifier_intel = SVC(**params).fit(x_train, y_train) from joblib import dump, load dump(classifier_intel, './model/svm_sklearnex.joblib')

I check the tow models, they are diferent.

How can I use speed up in the training step, and use as a normal sklearn model in inference step?
Thanks and best wishes!

Yan

[napetrov]

@gaosiy With current design there is no support general support for non-sklearnex inference. This is area that we would consider looking on, but not in nearest future.

Also it's important to note that skelearnex not necessary creates scikit-learn compatible models and this is part of optimizations that are used. So simple linear models 1:1 mapping is possible, while for more complex models such as SVM, Random Forest models would be different.