what metric used for classification? about autofeat HOT 4 CLOSED

I see from your code
rreg = LogisticRegression(class_weight="balanced")
so actually any
parameter from
https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LogisticRegression.html
class sklearn.linear_model.LogisticRegression(penalty='l2', dual=False, tol=0.0001, C=1.0, fit_intercept=True, intercept_scaling=1, class_weight=None, random_state=None, solver='lbfgs', max_iter=100, multi_class='auto', verbose=0, warm_start=False, n_jobs=None, l1_ratio=None)

can be used , for example
solver='liblinear', penalty='l1',

but only not C=3.0

???

def test_autofeat(dataset, feateng_steps=2):
# load data
X, y, units = load_classification_dataset(dataset)
# split in training and test parts
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=12)
# run autofeat
afreg = AutoFeatClassifier(verbose=1, feateng_steps=feateng_steps, units=units)
# fit autofeat on less data, otherwise ridge reg model with xval will overfit on new features
X_train_tr = afreg.fit_transform(X_train, y_train)
X_test_tr = afreg.transform(X_test)
print("autofeat new features:", len(afreg.new_feat_cols_))
print("autofeat Acc. on training data:", accuracy_score(y_train, afreg.predict(X_train_tr)))
print("autofeat Acc. on test data:", accuracy_score(y_test, afreg.predict(X_test_tr)))
# train rreg on transformed train split incl cross-validation for parameter selection
print("# Logistic Regression")
rreg = LogisticRegression(class_weight="balanced")
param_grid = {"C": np.logspace(-4, 4, 10)}

from autofeat.

Just take the features that are returned in the dataframe when calling the .fit_transform(X, y) function of the autofeat model and then train your own classification model with whatever parameters and metric you want.

from autofeat.

but features selection should be done for some specific metric to make it efficient for this metric
for example if features selected for accuracy then they can not be good for F1 metric?

from autofeat.

The features are not per se selected based on any metric but only based on the coefficients of the model. The models themselves are fitted according to their own optimization function (e.g. a linear regression model always minimizes the mean squared error, however, the scoring metric used to evaluate a trained model by default is R^2). The metric comes into play when selecting the hyperparameters (e.g. C) of the model in the internal cross-validation, however, this should not make a huge difference as the amount of regularization that is required is dependent more on how noisy your dataset is, not which metric you use to score it.

from autofeat.