def main():
X_train, X_test, y_train, y_test = prank_data_split(
'../dataset/ratings.csv', 0.2)
#cross validation#
depth_array = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11])
depth = kfoldcv(X_train.as_matrix(), y_train.as_matrix(), 5, depth_array,
'../dataset/movies.csv')
#prediction#
print('predict using decision tree with max depth', depth, ':')
X_train, X_test, y_train, y_test = generate_matrix(
X_train, X_test, y_train, y_test, '../dataset/movies.csv')
regr = decision_tree(X_train, y_train, depth)
y_predicted = predict(regr, X_test)
rmse = get_RMSE(y_test, y_predicted)
print('rmse:', rmse)
mae = get_MAE(y_test, y_predicted)
print('mae:', mae)
#specificity, sensitivity, precision, accuracy
spec, sens, prec, accu = get_spec_sens_prec_accu(y_test, y_predicted)
print('spec:', spec)
print('sens:', sens)
print('prec:', prec)
print('accu:', accu)
def performing_algorithm(X, y, X_test):
"""
:param X: Matrix
:param y: Matrix
:param X_test: Matrix
:return: Prediction of chosen algorithm
"""
if args.algorithm == "linear_regression":
return linear_regression(X, y, X_test)
elif args.algorithm == "decision_tree":
return decision_tree(X, y, X_test)
elif args.algorithm == "SVM":
return SVM(X, y, X_test)
def get_best_attribute(attributes, examples):
"""
workers = parallel.Workers()
workers.initialize_n_workers(4)
workers.set_function(get_attribute_info_gain)
workers.start()
inputs = [[a, examples] for a in attributes]
output = workers.run_over_data(inputs)
"""
best_gain = float("-inf")
best_attr = None
for a in attributes:
gain, attr = get_attribute_info_gain(a, examples)
if gain < best_gain:
best_gain = gain
best_attr = a
return best_attr, decision_tree(best_attr, dt_util.get_goal_counts(examples), dt_util.GOAL_INDEX)
def get_best_attribute(attributes, examples):
"""
workers = parallel.Workers()
workers.initialize_n_workers(4)
workers.set_function(get_attribute_info_gain)
workers.start()
inputs = [[a, examples] for a in attributes]
output = workers.run_over_data(inputs)
"""
best_gain = float("-inf")
best_attr = None
for a in attributes:
gain, attr = get_attribute_info_gain(a, examples)
if gain < best_gain:
best_gain = gain
best_attr = a
return best_attr, decision_tree(best_attr,
dt_util.get_goal_counts(examples),
dt_util.GOAL_INDEX)
def __init__(self, data, labels, num_trees, weights=None, randomized=False):
self.data = data
self.labels = labels
self.trees = []
num_obs = data.shape[0]
num_features = data.shape[1]
# uniformly weight data as default (can be modified for boosting)
if weights == None:
weights = 1.0/num_obs*numpy.ones((num_obs,1))
# split into num_trees training sets with data sampled with replacement according to weighting scheme
# the sets are the same size as the original training set
data_sets = numpy.zeros((num_trees, num_obs, num_features))
label_sets = numpy.zeros((num_trees, num_obs, 1))
for i in range(num_trees):
for j in range(num_obs):
sampled_obs_index = self.sample_index(weights)
data_sets[i,j] = data[sampled_obs_index, :]
label_sets[i,j] = labels[sampled_obs_index]
# train num_trees decision trees
for i in range(num_trees):
self.trees.append(decision_tree(data_sets[i,:,:], label_sets[i,:], randomized))
[90, 1, 0, 3.5], # FIT
[75, 1, 1, 3.1], # FIT
[85, 2, 1, 3.1], # NOT_FIT
[65, 0, 1, 2.1], # NOT_FIT
[70, 1, 0, 3.0]
]) # NOT_FIT
# класс для каждой кандидатки:
Y = np.array(
[FIT, FIT, FIT, FIT, NOT_FIT, FIT, FIT, NOT_FIT, NOT_FIT, NOT_FIT])
# типы переменных в столбцах обучающей выборки
scale = np.array([NUMERICAL, CATEGORICAL, CATEGORICAL, NUMERICAL])
# рекурсивное построение дерева решений
decision_tree(X, Y, scale)
# классификация каждого примера с помощью
# классификатора, созданного на основе дерева
y = np.array([clf.classify(X[i, :]) for i in range(len(X))])
# классификация успешна, если все примеры правильно классифицированы
if np.all(y == Y):
print('\nclassification success!\n')
else:
print('\nclassification fail... :(\n')
# проверка себя с помощью классификатора
# TODO: после того, как вы построили дерево решений и реализовали на его
# основе функцию classify, раскомментируйте код ниже и проверьте себя,
# подходите ли вы на роль ассистентки профессора Буковски :)
from decision_tree import *
from numpy import array
import scipy.io as sio
def max_depth(node):
if node.is_leaf():
return node.depth
else:
return max(max_depth(node.left_child), max_depth(node.right_child))
all_data = sio.loadmat('spam.mat')
dt = decision_tree(all_data['Xtrain'][0:3200,:], all_data['ytrain'][0:3200], True)
score = 0
for i in range(3200, 3450):
score += dt.classify(all_data['Xtrain'][i,:]) == all_data['ytrain'][i][0]
print 'score = ' + str(score)
print 'error = ', str(1 - float(score) / 250)
print 'max depth = ', max_depth(dt.root)
dt = decision_tree(all_data['Xtrain'], all_data['ytrain'], True)
score = 0
for i in range(3450):
score += dt.classify(all_data['Xtrain'][i,:]) == all_data['ytrain'][i][0]
print 'score = ' + str(score)
print 'error = ', str(1 - (float(score) / 3450))
print 'max depth = ', max_depth(dt.root)
#dt = decision_tree(all_data['Xtrain'], all_data['ytrain'])
