def prunning(class_weight='balanced'):
datasets = ['breast_cancer', 'kdd']
min_samples_leaf_list = [1, 25]
max_depth_list = [None, 4]
for dataset in datasets:
if dataset == 'kdd':
transform_data = True
random_slice = 10000
else:
transform_data = False
random_slice = None
x_train, x_test, y_train, y_test = data_service.load_and_split_data(scale_data=True,
transform_data=transform_data,
random_slice=random_slice,
random_seed=None,
dataset=dataset,
test_size=.5)
for min_samples_leaf in min_samples_leaf_list:
for max_depth in max_depth_list:
dt_learner = DTLearner(criterion='entropy', min_samples_leaf=min_samples_leaf, max_depth=max_depth,
class_weight=class_weight)
dt_accuracy_score, dt_fit_time, dt_predict_time = dt_learner.fit_predict_score(x_train, y_train,
x_test, y_test)
print('DT set: {0}, min_samples_leaf: {1}, max_depth: {2}, score: {3}, fit_time: {4}, predict_time: {5}'
.format(dataset, min_samples_leaf, max_depth, dt_accuracy_score, dt_fit_time, dt_predict_time))
dt_learner.draw_tree('-{0}-min_samples_leaf{1}-max_depth{2}'.format(dataset, min_samples_leaf,
max_depth))
def gini_vs_entropy():
min_samples_leaf = 1
max_depth = None
class_weight = 'balanced'
datasets = ['breast_cancer', 'kdd']
criterions = ['gini', 'entropy']
for dataset in datasets:
if dataset == 'kdd':
transform_data = True
random_slice = 10000
else:
transform_data = False
random_slice = None
x_train, x_test, y_train, y_test = data_service.load_and_split_data(scale_data=True,
transform_data=transform_data,
random_slice=random_slice,
random_seed=None,
dataset=dataset,
test_size=.5)
for criterion in criterions:
print(dataset, criterion)
dt_learner = DTLearner(criterion=criterion, min_samples_leaf=min_samples_leaf, max_depth=max_depth,
class_weight=class_weight)
dt_accuracy_score, dt_fit_time, dt_predict_time = dt_learner.fit_predict_score(x_train, y_train,
x_test, y_test)
print('DT set: {0}, criterion: {1} score: {2}, fit_time: {3}, predict_time: {4}, {2}'
.format(dataset, criterion, dt_accuracy_score, dt_fit_time, dt_predict_time))
dt_learner.draw_tree('-{0}-{1}'.format(dataset, criterion))
nn_solver = 'lbfgs'
#{'activation': 'relu', 'alpha': 0.0001, 'hidden_layer_sizes': (100,), 'learning_rate_init': 0.01, 'solver': 'lbfgs'}
nn_learner = NNLearner(hidden_layer_sizes=nn_hidden_layer_sizes,
max_iter=200,
solver=nn_solver,
activation=nn_activation,
alpha=alpha,
learning_rate=nn_learning_rate,
learning_rate_init=nn_learning_rate_init)
x_train, x_test, y_train, y_test = data_service.load_and_split_data(
scale_data=scale_data,
transform_data=transform_data,
random_slice=random_slice,
random_seed=random_seed,
dataset=dataset,
test_size=test_size)
nn_accuracy_score, nn_fit_time, nn_predict_time = nn_learner.fit_predict_score(
x_train, y_train, x_test, y_test)
print('nn: {0}, {1}, {2}'.format(nn_accuracy_score, nn_fit_time,
nn_predict_time))
print('---------------------------------------------------------------')
print(nn_learner.estimator.coefs_)
print('---------------------------------------------------------------')
print(nn_learner.estimator.coefs_[0].shape)
print(nn_learner.estimator.coefs_[1].shape)
print(nn_learner.estimator.n_outputs_)
print("Mean Absolute Error: ")
# mse = np.sqrt(np.mean(np.square(test_predictions - Y_test)))
mae = np.mean(np.abs(test_predictions - Y_test))
print(mae)
print("RMSE: ")
rmse = np.sqrt(np.mean(np.square(test_predictions - Y_test)))
print(rmse)
return mae
np.set_printoptions(suppress=True, precision=2)
X_train, X_test, Y_train, Y_test = \
data_service.load_and_split_data(scale_data=True, transform_data=False, test_size=0.2, random_slice=None,
random_seed=None, dataset='boston')
mean_absolute_errors = []
max_depths = range(1, 10)
for max_depth in max_depths:
#do it 5 times for the depth
mean_abs_errors_for_depth = []
for train_run in range(5):
X_train, X_test, Y_train, Y_test = \
data_service.load_and_split_data(scale_data=True, transform_data=False, test_size=0.2, random_slice=None,
random_seed=None, dataset='boston')
mean_abs_error = fit_predict_score(max_depth, X_train, X_test, Y_train,
Y_test)
mean_abs_errors_for_depth.append(mean_abs_error)