def knn_train(dimensions, train_count, test_count, iterations=10):
# For evaluation
BCE_list_train = []
BCE_list_test = []
for i in range(iterations):
# File import
train_x_data, train_y_data = create_toy_data(dimensions,
train_count,
test_count,
add_outliers=True,
training=True)
test_x_data, test_y_data = create_toy_data(dimensions,
train_count,
test_count,
add_outliers=True,
training=False)
# Feature engineering
X_train = train_x_data
X_test = test_x_data
# Model
knn = kNN()
# Training (Learning)
knn.fit(X_train, train_y_data)
# Predicting a training set
y_hat_train = knn.predict(X_train)
# Predicting a test set
y_hat_test = knn.predict(X_test)
# Evaluating
BCE_value_train = binary_cross_entropy(train_y_data, y_hat_train)
BCE_value_test = binary_cross_entropy(test_y_data, y_hat_test)
# Appending
BCE_list_train.append(BCE_value_train)
BCE_list_test.append(BCE_value_test)
# Plotting
if dimensions == 2 and i == 0:
x1_test, x2_test = np.meshgrid(np.linspace(-5, 15, 100),
np.linspace(-5, 15, 100))
X_test_plot = np.array([x1_test, x2_test]).reshape(2, -1).T
# logistic regression
y_hat_plot = knn.predict(X_test_plot)
plot(train_x_data, train_y_data, test_x_data, test_y_data, x1_test,
x2_test, y_hat_plot, "./Results/knn_result")
# average MSE
(average_BCE_train, BCE_std_train) = average_metric(BCE_list_train)
(average_BCE_test, BCE_std_test) = average_metric(BCE_list_test)
return (average_BCE_train, BCE_std_train), (average_BCE_test, BCE_std_test)
import numpy as np
import random
from App.Pre_processing.data_generation import create_toy_data
if __name__ == "__main__":
for i in range(10):
if i < 5:
train_x_data, train_y_data = create_toy_data()
test_x_data, test_y_data = create_toy_data(training=False)
else:
train_x_data, train_y_data = create_toy_data(add_outliers=True)
test_x_data, test_y_data = create_toy_data(add_outliers=True,
training=False)
np.savetxt("./Data/x_{}.csv".format(i), train_x_data)
np.savetxt("./Data/y_{}.csv".format(i), train_y_data)
np.savetxt("./Data/test_x_{}.csv".format(i), test_x_data)
np.savetxt("./Data/test_y_{}.csv".format(i), test_y_data)
def decision_tree_train(dimensions, train_count, test_count, iterations=10):
# Set Hyperparameters
max_depth = 5
min_size = 10
# For evaluation
BCE_list_train = []
BCE_list_test = []
Terminal_count_list = []
for i in range(iterations):
# File import
train_x_data, train_y_data = create_toy_data(dimensions, train_count, test_count, add_outliers=True, training=True)
test_x_data, test_y_data = create_toy_data(dimensions, train_count, test_count, add_outliers=True, training=False)
# Feature engineering
X_train = train_x_data
X_test = test_x_data
# Model
decision_tree = Decision_tree()
# combine the features and targets
train_y_data_tree = train_y_data[:, None]
Tree_input = np.hstack((X_train, train_y_data_tree))
# Training (Learning)
tree, terminal_count = decision_tree.build_tree(Tree_input, max_depth, min_size, np.shape(Tree_input)[1]-1)
# Predicting a training set
y_hat_train = decision_tree.predicts(tree, X_train)
# Predicting a test set
y_hat_test = decision_tree.predicts(tree, X_test)
# Evaluating
BCE_value_train = binary_cross_entropy(train_y_data, y_hat_train)
BCE_value_test = binary_cross_entropy(test_y_data, y_hat_test)
# Appending
BCE_list_train.append(BCE_value_train)
BCE_list_test.append(BCE_value_test)
Terminal_count_list.append(terminal_count)
# Plotting
if dimensions == 2 and i == 0:
x1_test, x2_test = np.meshgrid(np.linspace(-5, 15, 100), np.linspace(-5, 15, 100))
X_test_plot = np.array([x1_test, x2_test]).reshape(2, -1).T
# logistic regression
y_hat_plot = decision_tree.predicts(tree, X_test_plot)
plot(train_x_data, train_y_data, test_x_data, test_y_data, x1_test, x2_test, y_hat_plot, "./Results/tree_result")
# average MSE
(average_BCE_train, BCE_std_train) = average_metric(BCE_list_train)
(average_BCE_test, BCE_std_test) = average_metric(BCE_list_test)
average_terminal_count = np.average(Terminal_count_list)
return (average_BCE_train, BCE_std_train), (average_BCE_test, BCE_std_test), average_terminal_count