All geek questions in one place

Correlation between hyperparameters of classifiers

I am wondering if there is some correlation between the hyperparameters of two different classifiers.

For example: let's say that we run LogisticRegressionin the data set with the best hyperparameters (by searching through GridSearch) and want to run another classifier of the type SVC( SVMclassifier) ​​in one data set, but instead of finding all the hyperparameters with GridSearch, can we fix some values ​​(or reduce the range to limit the search space for GridSearch) hyperparameters?

As an experiment, I used qualifiers scikit-learnsuch as LogisticRegression, SVS, LinearSVC, SGDClassifierand Perceptronto classify some well-known data sets. In some cases, I can see some correlation empirically, but not always for all data sets.

So please help me clarify this point.

+4
source share
1 answer

, . , -, . , SVC rbf sigmoid. , rbf intercept parameter C, 0,001, sigmoid kernel over the same data may fit with C` 0.00001. . . , :

, .

. . .

, , a, , , a.

+4

Source: https://habr.com/ru/post/1685786/

More articles:


All Articles