def train_and_validate(algorithm):
"""This function is mandated by the requirements. Argument
'algorithm' takes values of 'naive_bayes', 'decision_tree', 'knn',
and 'svm', and the function will train the respective classifier
on training data, predict on the testing data, and print that
classifier's testing accuracy.
"""
# I'm going to assume we should use the specified classifier by
# itself, and using it as the base classifier in an ensemble is
# cheating.
#
# Below maps algorithm name to a function producing a model
# from training data:
algos = {
"naive_bayes": train_naive_bayes,
"decision_tree": train_decision_tree,
"knn": train_knn,
"svm": train_svm,
}
# Load data and train respective model:
train_X, train_y, test_X, test_y = data_preprocessing.get_processed_data()
train_fn = algos[algorithm]
model = train_fn(train_X, train_y)
# Get testing accuracy:
predict_y = model.predict(test_X)
test_acc = sklearn.metrics.accuracy_score(test_y, predict_y)
print(test_acc)
break
print("Feature \"{1}\" raises accuracy from {2} to {3}".format(
i, best_feature, base_accuracy, best_accuracy))
base_accuracy = best_accuracy
best_features.append(best_feature)
incr_accuracy.append(best_accuracy)
print("{0} features: {1}".format(len(best_features), best_features))
features.remove(best_feature)
df = pd.DataFrame.from_dict({
"Feature": best_features,
"Accuracy": incr_accuracy,
})
return df
train_X, train_y, test_X, test_y = data_preprocessing.get_processed_data()
print("-" * 70)
print("kNN:")
print("-" * 70)
knn = sklearn.neighbors.KNeighborsClassifier(n_jobs=-1)
params = {'n_neighbors': range(1, 17), 'weights': ['distance', 'uniform']}
features = [
'education_num', 'marital_status_Married-civ-spouse', 'net_capital'
]
clf = sklearn.model_selection.GridSearchCV(knn, params)
clf.fit(train_X[features], train_y)
print(clf.cv_results_)
print("Best score: {0}".format(clf.best_score_))
print("Best params: {0}".format(clf.best_params_))