easy-gscv 0.1.1

A high level library gridsearch / cross evaluation library for scikit-learn

Easy Grid Search / Cross Validation

From data to score in 4 lines of code.

This library allows you to quickly train machine learning classifiers by automatically splitting the dataset and using both grid search and cross validation in the training process. Users can either pass define the parameters themselves or let the GSCV object choose them automatically (based on the classifier).

This library is an extension of the scikit-learn project.

View on pypi

Example:

from sklearn.neural_network import MLPClassifier
from sklearn import datasets
from easy_gscv.classifiers import GSCV

# Create test dataset
iris = datasets.load_iris()
X = iris.data
y = iris.target
clf = MLPClassifier()

# Create model instance
gscv_model = GSCV(clf(), X, y)

# Get score
gscv_model.score()

install

requires python 3.7+

pip install easy-gscv

create

from easy_gscv.models import GSCV
clf = LogisticRegression()
gscv_model = GSCV(
    clf(), X, y, cv=15, n_jobs=-1, params={
        'C': [10, 100],
        'penalty': ['l2']
    }
)

No need to create separate train / test datasets, the model does this automatically on initialization. If no parameters are provided the grid search is performed on a default set. But these can be overridden.

The number of folds to be used for cross validation can be specified by using the cv keyword. To speed up the training process you can use the n_jobs parameter to set the number of cpu cores to use (or set it to -1 to use all available.)

The model accepts either sklearn classifiers or string values. You can get a list of valid classifiers by calling the 'classifiers' property. Passing string arguments to the GSCV object in turn saves you from having to import sklearn classifiers yourself.

gscv_model = GSCV('RandomForestClassifier',, X, y)
gscv_model.classifiers

'KNeighborsClassifier',
'RandomForestClassifier',
'GradientBoostingClassifier',
'MLPClassifier',
'LogisticRegression',

score

gscv_model.score()

The grid search is performed on the training data. Use the score method to evaluate how well the model can be generalized by scoring it against the test dataset.

get_best_estimator

gscv_model.get_best_estimator()

Returns the best scoring sklearn classifier (based on training data). As its a valid scikit-learn classifier, you can use it do anything that you could do with sklearn classifier.

The following classifiers are currently supported. With the eventual goal of supporting all scikit-learn classifiers in the future.

• KNeighborsClassifier
• RandomForestClassifier
• GradientBoostingClassifier
• MLPClassifier
• LogisticRegression

get_fit_details

As cross validation returns an average, it can be helpful to get a more detailed overview of the best scoring classifier.

This method returns a table like the one displayed below, which then can be used to further refine the choice or parameters for subsequent runs.

clf = KNeighborsClassifier()
gscv_model = GSCV(clf(), X, y)
gscv_model.get_fit_details()

0.965 (+/-0.026) for {'weights': 'uniform', 'n_neighbors': 3}
0.977 (+/-0.013) for {'weights': 'distance', 'n_neighbors': 3}
0.979 (+/-0.011) for {'weights': 'uniform', 'n_neighbors': 5}
0.979 (+/-0.011) for {'weights': 'distance', 'n_neighbors': 5}
0.976 (+/-0.018) for {'weights': 'uniform', 'n_neighbors': 8}
0.975 (+/-0.018) for {'weights': 'distance', 'n_neighbors': 8}
0.971 (+/-0.022) for {'weights': 'uniform', 'n_neighbors': 12}
0.973 (+/-0.024) for {'weights': 'distance', 'n_neighbors': 12}
0.973 (+/-0.025) for {'weights': 'uniform', 'n_neighbors': 15}