def main():
settings_path = get_settings_path_from_arg("decision_tree")
settings = load_settings(settings_path)
init_logger(settings.log.dir, "decision_tree_classifier", logging.DEBUG)
dl = DataLoader()
dl.load_train(settings.data.train_path)
dl.load_test(settings.data.test_path)
ms = ModelSelector()
# Load test and training
dl = DataLoader()
dl.load_train(settings.data.train_path)
dl.load_test(settings.data.test_path)
dl.scale_data()
# Image feature extraction
k = np.size(dl.classes) *10
dl.load_from_images(settings.data.image_path, k, k*3, verbose=False)
# Add Data Selector
ds = DataSelector(
dl.id_train, dl.x_train, dl.y_train,
dl.id_test, dl.x_test
)
ds.add_all()
# Use lasso
ds.auto_remove_lasso(0.17)
# Dimensionality reduction
dr = DataReducer(ds.train_x, ds.test_x)
dr.pca_data_reduction()
ds = DataSelector(
dl.id_train, dr.x_train, dl.y_train,
dl.id_test, dr.x_test
)
ds.add_all()
# Add data selection to model selector
ms.add_selector("all_feature", ds)
clf = DecisionTreeClassifier()
ms.add_classifier("", clf)
ms.get_best_model(k=10, plot=True)
ms.generate_submission(settings.data.submission_dir, dl.classes)
def main():
settings_path = get_settings_path_from_arg("qda_classifier")
settings = load_settings(settings_path)
init_logger(settings.log.dir, "qda_classifier", logging.DEBUG)
ms = ModelSelector()
# Load test and training
dl = DataLoader()
dl.load_train(settings.data.train_path)
dl.load_test(settings.data.test_path)
dl.scale_data()
# Image feature extraction
k = np.size(dl.classes) * 10
dl.load_from_images(settings.data.image_path, k, k * 3, verbose=False)
# Add Data Selector
ds = DataSelector(dl.id_train, dl.x_train, dl.y_train, dl.id_test,
dl.x_test)
ds.add_all()
# Use lasso
ds.auto_remove_lasso(0.17)
# Dimensionality reduction
dr = DataReducer(ds.train_x, ds.test_x)
dr.pca_data_reduction()
ds = DataSelector(dl.id_train, dr.x_train, dl.y_train, dl.id_test,
dr.x_test)
ds.add_all()
# Add Classifier to model selector
clf = QuadraticDiscriminantAnalysis(store_covariance=True)
ms.add_classifier("store_covariance=True", clf)
clf = QuadraticDiscriminantAnalysis(store_covariance=False)
ms.add_classifier("store_covariance=False", clf)
clf = QuadraticDiscriminantAnalysis(priors=None,
reg_param=0.0,
store_covariance=False,
store_covariances=None,
tol=0.0001)
ms.add_classifier("final", clf)
# Get best model
ms.get_best_model(k=10, plot=True)
ms.generate_submission(settings.data.submission_dir, dl.classes)
def main():
settings_path = get_settings_path_from_arg("naive_bayes_classifier")
settings = load_settings(settings_path)
init_logger(settings.log.dir, "naive_bayes_classifier", logging.DEBUG)
ms = ModelSelector()
# Load test and training
dl = DataLoader()
dl.load_train(settings.data.train_path)
dl.load_test(settings.data.test_path)
dl.scale_data()
# Add Data Selector
ds = DataSelector(dl.id_train, dl.x_train, dl.y_train, dl.id_test,
dl.x_test)
ds.add_all()
# Image feature extraction
k = np.size(dl.classes) * 10
dl.load_from_images(settings.data.image_path, k, k * 3, verbose=False)
# Add Data Selector
ds = DataSelector(dl.id_train, dl.x_train, dl.y_train, dl.id_test,
dl.x_test)
ds.add_all()
# Use lasso
ds.auto_remove_lasso(0.17)
# Dimensionality reduction
dr = DataReducer(ds.train_x, ds.test_x)
dr.pca_data_reduction()
ds = DataSelector(dl.id_train, dr.x_train, dl.y_train, dl.id_test,
dr.x_test)
ds.add_all()
# Add data selection to model selector
ms.add_selector("all_feature", ds)
# Add classifier to model selector
clf = BernoulliNB(alpha=3.5, binarize=0.03)
ms.add_classifier("bernoulli", clf)
# Get best model
ms.get_best_model(k=10)
ms.generate_submission(settings.data.submission_dir, dl.classes)
def main():
settings_path = get_settings_path_from_arg("random_forest_classifier")
settings = load_settings(settings_path)
init_logger(settings.log.dir, "random_forest_classifier", logging.DEBUG)
ms = ModelSelector()
# Load test and training
dl = DataLoader()
dl.load_train(settings.data.train_path)
dl.load_test(settings.data.test_path)
dl.scale_data()
# Image feature extraction
k = np.size(dl.classes) * 10
dl.load_from_images(settings.data.image_path, k, k * 3, verbose=False)
# Add Data Selector
ds = DataSelector(dl.id_train, dl.x_train, dl.y_train, dl.id_test,
dl.x_test)
ds.add_all()
# Use lasso
ds.auto_remove_lasso(0.17)
# Dimensionality reduction
dr = DataReducer(ds.train_x, ds.test_x)
dr.pca_data_reduction()
ds = DataSelector(dl.id_train, dr.x_train, dl.y_train, dl.id_test,
dr.x_test)
ds.add_all()
# Add data selection to model selector
ms.add_selector("all_feature", ds)
# Add Classifier to model selector
clf = RandomForestClassifier(n_estimators=13,
random_state=0,
min_samples_leaf=3,
bootstrap=False)
ms.add_classifier(
"n_estimators=13, random_state=0, min_samples_leaf=3, bootsrap=False",
clf)
# Get best model
ms.get_best_model(k=10)
ms.generate_submission(settings.data.submission_dir, dl.classes)
def main():
settings_path = get_settings_path_from_arg("k_neighbors_classifier")
settings = load_settings(settings_path)
init_logger(settings.log.dir, "k_neighbors_classifier", logging.DEBUG)
ms = ModelSelector()
# Load test and training
dl = DataLoader()
dl.load_train(settings.data.train_path)
dl.load_test(settings.data.test_path)
dl.scale_data()
# Image feature extraction
k = np.size(dl.classes) * 10
dl.load_from_images(settings.data.image_path, k, k * 3, verbose=False)
# Add Data Selector
ds = DataSelector(dl.id_train, dl.x_train, dl.y_train, dl.id_test,
dl.x_test)
ds.add_all()
# Use lasso
ds.auto_remove_lasso(0.17)
# Dimensionality reduction
dr = DataReducer(ds.train_x, ds.test_x)
dr.pca_data_reduction()
ds = DataSelector(dl.id_train, dr.x_train, dl.y_train, dl.id_test,
dr.x_test)
ds.add_all()
# Add data selection to model selector
ms.add_selector("all_feature", ds)
for i in range(1, 40):
clf_k = KNeighborsClassifier(i, weights="distance", p=1)
ms.add_classifier("k_{}_distance_p1".format(i), clf_k)
# Get best model
ms.get_best_model(k=10, plot=True)
ms.generate_submission(settings.data.submission_dir, dl.classes)
logging.info("[ImageFeatureExtractor] Histogram generated")
logging.debug("[ImageFeatureExtractor] Histogram:{}".format(histo_list))
return histo_list
def extract_descriptor_to_csv(img_path):
"""Extract image descriptor into csv"""
def extract_feature_to_csv(img_path, ids, k, batch_size=None, verbose=False):
"""Extract feature from image and save to csv"""
if __name__ == "__main__":
settings_path = get_settings_path_from_arg("image_feature_extractor")
settings = load_settings(settings_path)
init_logger(settings.log.dir, "image_feature_extractor", logging.DEBUG)
train_data = pd.read_csv(settings.data.train_path)
test_data = pd.read_csv(settings.data.test_path)
species = train_data["species"]
train_data = train_data.drop("species", axis=1)
k = np.size(species) * 10
batch_size = np.size(os.listdir(settings.data.image_path)) * 3
print(
get_feature(settings.data.image_path,
test_data['id'],
k,
def main():
settings_path = get_settings_path_from_arg("main_classifier")
settings = load_settings(settings_path)
init_logger(settings.log.dir, "main_classifier", logging.DEBUG)
ms = ModelSelector()
# Load test and training data
dl = DataLoader()
dl.load_train(settings.data.train_path)
dl.load_test(settings.data.test_path)
dl.scale_data()
# Image feature extraction
k = np.size(dl.classes) * 10
dl.load_from_images(settings.data.image_path, k, k * 3, verbose=False)
# Add Data Selector
ds = DataSelector(dl.id_train, dl.x_train, dl.y_train, dl.id_test,
dl.x_test)
ds.add_all()
# Use lasso
ds.auto_remove_lasso(0.17)
# Dimensionality reduction
dr = DataReducer(ds.train_x, ds.test_x)
dr.pca_data_reduction()
ds = DataSelector(dl.id_train, dr.x_train, dl.y_train, dl.id_test,
dr.x_test)
ds.add_all()
# Add data selection to model selector
ms.add_selector("all_feature", ds)
# Instantiate all the classifiers to be added
clf_knn = KNeighborsClassifier(6, weights="distance", p=1)
clf_adaboost = AdaBoostClassifier(learning_rate=0.01)
clf_dectree = DecisionTreeClassifier(min_impurity_decrease=0.02)
clf_gaussian = GaussianNB()
clf_lda = LinearDiscriminantAnalysis()
clf_gradientboost = GradientBoostingClassifier(n_estimators=100,
random_state=0,
min_samples_leaf=3,
verbose=True)
clf_bernoullinb = BernoulliNB(alpha=3.5, binarize=0.03)
clf_nnkeras = NNKeras(ds.selected_x_test.shape[1])
clf_svc = SVC(probability=True, C=1000, gamma=1)
clf_nusvc = NuSVC(nu=0.1, gamma=10, probability=True)
clf_qda = QuadraticDiscriminantAnalysis(store_covariance=True)
clf_randomforest = RandomForestClassifier(n_estimators=13,
random_state=0,
min_samples_leaf=3,
bootstrap=False)
# Add all the classifiers to the model selector
ms.add_classifier("KNN", clf_knn)
ms.add_classifier("AdaBoost", clf_adaboost)
ms.add_classifier("Decision Tree", clf_dectree)
ms.add_classifier("Gaussian NB", clf_gaussian)
ms.add_classifier("LDA", clf_lda)
ms.add_classifier("Gradient Boosting", clf_gradientboost)
ms.add_classifier("Bernoulli NB", clf_bernoullinb)
ms.add_classifier("NN Keras", clf_nnkeras)
ms.add_classifier("SVC", clf_svc)
ms.add_classifier("NuSVC", clf_nusvc)
ms.add_classifier("QDA", clf_qda)
ms.add_classifier("Random Forest", clf_randomforest)
# Get best model
ms.get_best_model(k=10, plot=True)
self.pca_x_train = data_pca[:len(self.x_train)]
self.pca_x_train = pd.DataFrame(self.pca_x_train, columns=col_names)
self.pca_x_test = data_pca[len(self.x_train):]
self.pca_x_test = pd.DataFrame(self.pca_x_test, columns=col_names)
def get_pca_x_train(self):
""" get x_train after pca """
return self.pca_x_train
def get_pca_x_test(self):
""" get x_test after pca """
return self.pca_x_test
if __name__ == "__main__":
settings_path = get_settings_path_from_arg("data_reducer")
settings = load_settings(settings_path)
init_logger(settings.log.dir, "data_reducer", logging.DEBUG)
train_data = pd.read_csv(settings.data.train_path)
test_data = pd.read_csv(settings.data.test_path)
train_data = train_data.drop("species", axis=1)
dr = DataReducer(train_data, test_data)
dr.pca_data_reduction()
pca_train_data = dr.get_pca_x_train()
pca_test_data = dr.get_pca_x_test()
