def k_nearest_neighbours(self):
"""
for knn, i train on the training data using different :
1) n_neighbors,
2) weights
:return: test accuracy of the knn best model
"""
# define parameters
# n_neighbors = np.logspace(start=2, stop=6, base=2, num=5, dtype=np.int)
# weights = ('distance', 'uniform')
# best result over all n_neighbors: 32
# best result over all weights: 'distance'
# scale down parameters around its best result
np.random.seed(0)
n_neighbors = norm.rvs(loc=32, scale=10, size=3).astype(np.int)
weights = ('distance', 'uniform')
# get the best validated model
knn = K_nearest_neighbours(x_train=self.x_train,
y_train=self.y_train,
cv=5,
n_neighbors=n_neighbors,
weights=weights,
grid_search=True)
# print all possible parameter values and the best parameters
# knn.print_parameter_candidates()
# knn.print_best_estimator()
# return the accuracy score
return (knn.evaluate(data=self.x_train, targets=self.y_train),
knn.evaluate(data=self.x_test, targets=self.y_test))
def k_nearest_neighbours(self):
weights = ['uniform', 'distance']
n_neighbors = range(3, 15)
knn = K_nearest_neighbours(x_train=self.x_train,
y_train=self.y_train,
cv=3,
n_neighbors=n_neighbors,
weights=weights,
grid_search=True)
#knn.print_parameter_candidates()
#knn.print_best_estimator()
return (knn.evaluate(data=self.x_train, targets=self.y_train),
knn.evaluate(data=self.x_test, targets=self.y_test))
def k_nearest_neighbours(self):
n_neighbors = range(1, 100)
knn = K_nearest_neighbours(x_train=self.x_train,
y_train=self.y_train,
cv=3,
n_neighbors=n_neighbors,
random_search=True)
# knn.print_parameter_candidates()
# knn.print_best_estimator()
return (knn.evaluate(data=self.x_train,
targets=self.y_train,
average='micro'),
knn.evaluate(data=self.x_test,
targets=self.y_test,
average='micro'))
def k_nearest_neighbours(self):
"""
for knn, i train on the training data using different :
1) n_neighbors
2) weights
:return: ((accuracy_train, recall_train, precision_train),
(accuracy_test, recall_test, precision_test))
"""
# define parameters
# n_neighbors = np.logspace(start=1, stop=9, base=2, num=9, dtype=np.int)
# weights = ('distance', 'uniform')
# best result over all n_neighbors: 64
# best result over all weights: 'distance'
# scale down parameters around its best result (1st)
# n_neighbors = norm.rvs(loc=64, scale=32, size=20).astype(np.int)
# weights = ('distance', 'uniform')
# best result over all n_neighbors: 69
# best result over all weights: 'distance'
# scale down parameters around its best result (2nd)
scale = 5
n_neighbors = np.arange(start=69 - scale,
stop=69 + scale,
step=1,
dtype=np.int)
weights = ('distance', 'uniform')
# get the best validated model
knn = K_nearest_neighbours(x_train=self.x_train,
y_train=self.y_train,
cv=5,
n_neighbors=n_neighbors,
weights=weights,
grid_search=True)
# print all possible parameter values and the best parameters
# knn.print_parameter_candidates()
# knn.print_best_estimator()
return (knn.evaluate(data=self.x_train, targets=self.y_train),
knn.evaluate(data=self.x_test, targets=self.y_test))
def k_nearest_neighbours(self):
n_neighbors = range(1, 100, 1) # [1, 3, 5, ..., 99]
knn = K_nearest_neighbours(x_train=self.x_train,
y_train=self.y_train,
cv=3,
n_jobs=-1,
n_neighbors=n_neighbors,
grid_search=True)
knn.print_parameter_candidates()
knn.print_best_estimator()
return (knn.evaluate(data=self.x_train, targets=self.y_train),
knn.evaluate(data=self.x_test, targets=self.y_test))
def k_nearest_neighbours(self):
# define arguments given to GridSearchCV
n_neighbors = range(1, 100, 1) # [1, 3, 5, ..., 99]
# get the best validated model
knn = K_nearest_neighbours(x_train=self.x_train,
y_train=self.y_train,
cv=23,
n_jobs=-1,
n_neighbors=n_neighbors,
grid_search=True)
# print all possible parameter values and the best parameters
# knn.print_parameter_candidates()
# knn.print_best_estimator()
# return the accuracy score
return (knn.evaluate(data=self.x_train,
targets=self.y_train,
average='macro'),
knn.evaluate(data=self.x_test,
targets=self.y_test,
average='macro'))
def k_nearest_neighbours(self):
# Best K = 27
# Test Error: 0.802764
n_neighbors = range(1, 100)
# get the best validated model
knn = K_nearest_neighbours(x_train=self.x_train,
y_train=self.y_train,
cv=3,
n_neighbors=n_neighbors,
random_search=True)
# print all possible parameter values and the best parameters
knn.print_parameter_candidates()
knn.print_best_estimator()
# return the accuracy score
return knn.accuracy_score(x_test=self.x_test, y_test=self.y_test)