Auto-ML for only fixed estimator

Question

Am working on a binary classification with 1000 rows and 28 columns.

I would wish to use an Auto-ML solution to try out different combinations of hyperparameters etc but the algo should only be logistic regression.

I don't wish to use other algorithms for the lack of interpretability.

So, I would like my auto-ML solution to stick to logistic regression and try out different values for hyperparameters. Of course, I might use fixed estimators like Decision trees, random forests etc as well

Is there any auto-ML solution that can use fixed estimator?

I read about Tpot, Evalml, AutoML etc but they all try multiple algorithms and finally output the best one (which may not be logistic regression). How can I restrict my auto-ML solution to only use logistic regression?

Answer 1

You do not need an automatic machine learning (Auto-ML) solution to find the best hyperparameters for logistic regression. You can use grid search or random search.

Answer 2

I found that we can do this using Tpot's config_dict and pass this as input to the classifier function like as shown below

tpot_config = {
    'sklearn.linear_model.LogisticRegression': {
        'penalty': ["l1", "l2"],
        'C': [1e-4, 1e-3, 1e-2, 1e-1, 0.5, 1., 5., 10., 15., 20., 25.],
        'dual': [True, False]
    },
}

tpot = TPOTClassifier(max_time_mins=10,verbosity=2,
                      config_dict=tpot_config,scoring='f1')
tpot.fit(ord_train_t, y_train)

This will ensure that TPOT searches best pipeline based on the configs provided by the config_dict.

However, if there are any other ML tool, I am interested to know from others here as well