def xgb_params():
para_dict = {
"max_depth": rd(3, 10),
"learning_rate": uni(loc=0, scale=1),
"n_estimators": rd(50, 200),
'objective': ['reg:linear'],
}
return para_dict
def uniform(lower, upper):
return uni(lower, upper - lower)
def __init__(self, data_set_file, lowercase=False, use_idf=False,
developers_dict_file=None,
developers_list_file=None):
super().__init__(developers_dict_file, developers_list_file)
np.random.seed(0) # We set the seed
self.lowercase = lowercase
self.use_idf = use_idf
self._estimators = [("count", CountVectorizer( \
lowercase=self.lowercase, token_pattern=u"(?u)\S+")), \
("tf_idf", TfidfTransformer(use_idf=self.use_idf, smooth_idf=False))]
self._nearest_centroid_estimators = self._estimators + \
[("clf", NearestCentroid())]
self._naive_bayes_estimators = self._estimators + \
[("clf", MultinomialNB())]
self._linear_svm_estimators = self._estimators + \
[("clf", LinearSVC())]
self._logistic_regression_estimators = self._estimators + \
[("clf", LogisticRegression())]
self._perceptron_estimators = self._estimators + \
[("clf", Perceptron())]
self._stochastic_gradient_descent_estimators = \
self._estimators + [("clf", SGDClassifier())]
self._nearest_centroid_estimators_params = \
dict( \
clf__metric=["manhattan", "euclidean"] \
)
self._naive_bayes_estimators_params = \
dict( \
clf__alpha=np.linspace(0, 1, 11), \
clf__fit_prior=[True, False] \
)
self._linear_svm_estimators_params = \
dict( \
clf__C=np.logspace(-4, 4, 10), \
clf__loss=["squared_hinge", "hinge"], \
clf__class_weight=["balanced"] \
)
self._primal_logistic_regression_estimators_params = \
dict( \
clf__dual=[False],
clf__C=np.logspace(-4, 4, 10), \
clf__class_weight=["balanced"], \
clf__solver=["newton-cg", "sag", "lbfgs"], \
clf__multi_class=["multinomial"]
)
self._dual_logistic_regression_estimators_params = \
dict( \
clf__dual=[True],
clf__C=np.logspace(-4, 4, 10), \
clf__class_weight=["balanced"], \
clf__solver=["liblinear"], \
clf__multi_class=["ovr"]
)
self._perceptron_with_penalty_estimators_params = \
dict( \
clf__penalty=["l2", "elasticnet"],
clf__alpha=10.0**-np.arange(1,7), \
clf__class_weight=["balanced"]
)
self._perceptron_without_penalty_estimators_params = \
dict( \
clf__penalty=[None],
clf__class_weight=["balanced"]
)
self._stochastic_gradient_descent_estimators_params = \
dict( \
clf__loss=["hinge", "log", "modified_huber", \
"squared_hinge", "perceptron"], \
clf__penalty=["l2", "elasticnet"], \
clf__alpha=10.0**-np.arange(1,7), \
clf__class_weight=["balanced"], \
clf__average=[True, False]
)
# Below, there is a dictionary to store the names, the
# classifiers used, the parameters sent to the constructor of
# the classifiers and the fitted classifiers (grid search)
self._models_cv = { \
"NearestCentroid": [GridSearchCV, {
"estimator": Pipeline(self._nearest_centroid_estimators),
"param_grid": self._nearest_centroid_estimators_params,
"n_jobs": 8,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"MultinomialNB": [GridSearchCV, {
"estimator": Pipeline(self._naive_bayes_estimators),
"param_grid": self._naive_bayes_estimators_params,
"n_jobs": 1,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"LinearSVC": [GridSearchCV, {
"estimator": Pipeline(self._linear_svm_estimators),
"param_grid": self._linear_svm_estimators_params,
"n_jobs": 1,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"Primal LogisticRegression": [GridSearchCV, {
"estimator": Pipeline(self._logistic_regression_estimators),
"param_grid": self._primal_logistic_regression_estimators_params,
"n_jobs": 1,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"Dual LogisticRegression": [GridSearchCV, {
"estimator": Pipeline(self._logistic_regression_estimators),
"param_grid": self._dual_logistic_regression_estimators_params,
"n_jobs": 8,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"PerceptronWithPenalty": [GridSearchCV, {
"estimator": Pipeline(self._perceptron_estimators),
"param_grid": self._perceptron_with_penalty_estimators_params,
"n_jobs": 8,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"PerceptronWithoutPenalty": [GridSearchCV, {
"estimator": Pipeline(self._perceptron_estimators),
"param_grid": self._perceptron_without_penalty_estimators_params,
"n_jobs": 8,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"SGDClassifier": [GridSearchCV, {
"estimator": Pipeline(self._stochastic_gradient_descent_estimators),
"param_grid": self._stochastic_gradient_descent_estimators_params,
"n_jobs": 1,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
}
self._nearest_centroid_estimators_random_params = \
dict( \
clf__metric=["manhattan", "euclidean"] \
)
self._naive_bayes_estimators_random_params = \
dict( \
clf__alpha=uni(loc=0,scale=1), \
clf__fit_prior=[True, False] \
)
self._linear_svm_estimators_random_params = \
dict( \
clf__C=LogSpaceUniform(loc=-4,scale=8), \
clf__loss=["squared_hinge", "hinge"], \
clf__class_weight=["balanced"] \
)
self._primal_logistic_regression_estimators_random_params = \
dict( \
clf__dual=[False],
clf__C=LogSpaceUniform(loc=-4,scale=8), \
clf__class_weight=["balanced"], \
clf__solver=["newton-cg", "sag", "lbfgs"], \
clf__multi_class=["multinomial"]
)
self._dual_logistic_regression_estimators_random_params = \
dict( \
clf__dual=[True],
clf__C=LogSpaceUniform(loc=-4,scale=8), \
clf__class_weight=["balanced"], \
clf__solver=["liblinear"], \
clf__multi_class=["ovr"]
)
self._perceptron_with_penalty_estimators_random_params = \
dict( \
clf__penalty=["l2", "elasticnet"],
clf__alpha=LogSpaceUniform(loc=-6,scale=5), \
clf__class_weight=["balanced"]
)
self._perceptron_without_penalty_estimators_random_params = \
dict( \
clf__penalty=[None],
clf__class_weight=["balanced"]
)
self._stochastic_gradient_descent_estimators_random_params = \
dict( \
clf__loss=["hinge", "log", "modified_huber", \
"squared_hinge", "perceptron"], \
clf__penalty=["l2", "elasticnet"], \
clf__alpha=LogSpaceUniform(loc=-6,scale=5), \
clf__class_weight=["balanced"], \
clf__average=[True, False]
)
# Below, there is a dictionary to store the names, the
# classifiers used, the parameters sent to the constructor of
# the classifiers and the fitted classifiers (random search)
self._randomized_models_cv = { \
"NearestCentroid": [RandomizedSearchCV, {
"estimator": Pipeline(self._nearest_centroid_estimators),
"param_distributions": self._nearest_centroid_estimators_random_params,
"n_iter": 2,
"n_jobs": 8,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"random_state": 0,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"MultinomialNB": [RandomizedSearchCV, {
"estimator": Pipeline(self._naive_bayes_estimators),
"param_distributions": self._naive_bayes_estimators_random_params,
"n_iter": 22,
"n_jobs": 1,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"random_state": 0,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"LinearSVC": [RandomizedSearchCV, {
"estimator": Pipeline(self._linear_svm_estimators),
"param_distributions": self._linear_svm_estimators_random_params,
"n_iter": 20,
"n_jobs": 1,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"random_state": 0,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"Primal LogisticRegression": [RandomizedSearchCV, {
"estimator": Pipeline(self._logistic_regression_estimators),
"param_distributions": self._primal_logistic_regression_estimators_random_params,
"n_iter": 30,
"n_jobs": 1,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"random_state": 0,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"Dual LogisticRegression": [RandomizedSearchCV, {
"estimator": Pipeline(self._logistic_regression_estimators),
"param_distributions": self._dual_logistic_regression_estimators_random_params,
"n_iter": 10,
"n_jobs": 8,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"random_state": 0,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"PerceptronWithPenalty": [RandomizedSearchCV, {
"estimator": Pipeline(self._perceptron_estimators),
"param_distributions": self._perceptron_with_penalty_estimators_random_params,
"n_iter": 12,
"n_jobs": 8,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"random_state": 0,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"PerceptronWithoutPenalty": [RandomizedSearchCV, {
"estimator": Pipeline(self._perceptron_estimators),
"param_distributions": self._perceptron_without_penalty_estimators_random_params,
"n_iter": 1,
"n_jobs": 8,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"random_state": 0,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
"SGDClassifier": [RandomizedSearchCV, {
"estimator": Pipeline(self._stochastic_gradient_descent_estimators),
"param_distributions": self._stochastic_gradient_descent_estimators_random_params,
"n_iter": 120,
"n_jobs": 1,
"iid": False,
"cv": self._tscv,
"verbose": 10,
"random_state": 0,
"error_score": np.array([-1, -1]),
"scoring": accuracy_mrr_scoring_object
}, None], \
}
# Below, there is a dictionary to store the accuracy of each
# configuration on the test set (grid search)
self._configurations_accuracies = {}
# Below, there is a dictionary to store the MRR value of each
# configuration on the test set (grid search)
self._configurations_mrr_values = {}
# Below, there is a dictionary to store the accuracy of each
# configuration on the test set (random search)
self._randomized_configurations_accuracies = {}
# Below, there is a dictionary to store the MRR value of each
# configuration on the test set (random search)
self._randomized_configurations_mrr_values = {}
cleaned_results_file_name = "cleaned_tuning_" + \
"individual_classifier_generic_experiment_results.json"
self._cleaned_results_file_name = os.path.join( \
self._current_dir, cleaned_results_file_name)
self._data_set_file = os.path.join(self._current_dir, \
data_set_file)
log_file = os.path.join(self._current_dir, \
"tuning_individual_classifier_generic_experiment.log")
logging.basicConfig(filename=log_file, filemode="w", \
level=logging.DEBUG)
self._build_data_set()