break
mu = mu_new
return mu
# vectorized function
return np.apply_along_axis(unit_loc_estimator,
axis,
arr,
maxiter=maxiter,
tor=tor)
def fit(self, X, axis=0, maxiter=50, tor=0.001):
loc = self.loc_estimator(X, axis=axis, maxiter=maxiter, tor=tor)
scale = self.scale_estimator(X, axis=axis, maxiter=maxiter, tor=tor)
return loc, scale
if __name__ == '__main__':
m_est = MEstimator()
X_train, y_train, X_test, y_test = simulation_setup(n_i=1000,
n_o=200,
n_t=1000,
p=10,
sigma_e=0.25)
t1 = time.time()
m_estimated_scale = m_est.scale_estimator(X_train, maxiter=50)
m_estimated_loc = m_est.loc_estimator(X_train, maxiter=50)
standard_deviation = np.std(X_train, ddof=1, axis=0)
median = np.median(X_train, axis=0)
print('consumed time: %.5f s' % (time.time() - t1))
"""
computes accuracy of the model
:param X_test: ndarray, shape(n_samples, n_features)
Test data
:param y_test: ndarray, shape(n_samples,)
Labels of test data
:param prob_threshold: double, default: 0.5
probability threshold for determining the predicted labels
:return: double
accuracy of the logistic regression model.
"""
y_test = np.array(y_test, dtype=int)
y_predict = self.predict(X_test, prob_threshold=prob_threshold)
accuracy = np.mean(y_predict == y_test)
return accuracy
if __name__ == '__main__':
data_train, data_test, beta_actual = simulation_setup(n_i=1000, n_o=200, n_t=1000, p=10,
sigma_e=0.25)
X_train, y_train = data_train[:, :-1], data_train[:, -1]
X_test, y_test = data_test[:, :-1], data_test[:, -1]
classical_bootstrap = ClassicalBootstrap()
classical_bootstrap.fit(X_train, y_train)
print("classical bootstrap score: ", classical_bootstrap.score(X_test, y_test))
n_strata_arr = np.arange(2, 16, 2, dtype=int)
# lambda_ = np.arange(0.05, 0.55, 0.05)
n_i = 1000
# n_o_arr = (n_i * lambda_).astype(int)
score_strat_arr = []
# score_boot_arr = []
# for n_o in n_o_arr:
for n_strata in n_strata_arr:
score_strat = []
# score_boot = []
for i in range(10):
X_train, y_train, X_test, y_test = simulation_setup(n_i=n_i,
n_o=200,
n_t=n_i,
p=p)
# stratified
strat = StratifiedBootstrap()
strat.fit(X_train,
y_train,
n_bootstrap=5,
n_strata=n_strata,
fast=True)
score_strat.append(strat.score(X_test, y_test))
# Bootstrap
boot = ClassicalBootstrap()
boot.fit(X_train, y_train, n_bootstrap=5)
# score_boot.append(boot.score(X_test, y_test))
def predict(self, X_test):
if self.beta is None:
raise ValueError("MLE Model is not fitted yet")
# X_test = np.concatenate((np.ones(X_test.shape[0], 1), X_test), axis=1)
return self.probability(self.beta, X_test)
def accuracy(self, X_test, y_test, prob_threshold=0.5):
y_test = np.array(y_test, dtype=int)
y_predicted = (self.predict(X_test) >= prob_threshold).astype(int)
accuracy = np.mean(y_predicted == y_test)
return accuracy
if __name__ == '__main__':
data_train, data_test, beta_actual = simulation_setup(n_i=1000,
n_o=0,
n_t=1000,
p=20)
X_train, y_train = data_train[:, :-1], data_train[:, -1].astype(int)
X_test, y_test = data_test[:, :-1], data_test[:, -1].astype(int)
mle = MLE()
mle.fit(X_train, y_train)
acc = mle.accuracy(X_test, y_test)
print('accuracy:', acc)
print('fitted coefficients: \n', mle.beta)
print('actual coefficients: \n', beta_actual)
lr = LogisticRegression(fit_intercept=False, solver='lbfgs')
lr.fit(X_train, y_train)
score = lr.score(X_test, y_test)
print('LR accuracy: ', score)
print('LR coefficients: \n', lr.coef_[0])
print('RMSE: ', np.sqrt(np.mean((mle.beta - lr.coef_[0])**2)))
import numpy as np
import matplotlib.pyplot as plt
import matplotlib
matplotlib.use('MacOSX')
from rblr.influence_function_bootstrap import IFB
from rblr.classical_bootstrap import ClassicalBootstrap
from sklearn.linear_model import LogisticRegression
from rblr.simulation_setup import simulation_setup
matplotlib.rcParams['font.size'] = 10
q = np.linspace(0.1, 1, 20)
b = 10
X_train, y_train, X_test, y_test = simulation_setup(n_i=1000, n_o=200, p=8)
lr = LogisticRegression(solver='lbfgs')
lr.fit(X_train, y_train)
score_lr = lr.score(X_test, y_test)
class_boot = ClassicalBootstrap()
class_boot.fit(X_train, y_train, n_bootstrap=b)
score_class_boot = class_boot.score(X_test, y_test)
score_ifb = []
for q_ in q:
ifb = IFB(c=None, gamma=5)
ifb.fit(X_train, y_train, n_bootstrap=b, quantile_factor=q_)
score_ifb.append(ifb.score(X_test, y_test))
f1 = plt.figure(1, figsize=(7, 4.8))
plt.plot(q, score_ifb, label='IFB')
from rblr.simulation_setup import simulation_setup
from rblr.influence_function_bootstrap import IFB
from sklearn.metrics import precision_recall_fscore_support
from rblr.preprocessing import Preprocessor
import matplotlib.pyplot as plt
n_i = 1000
n_o = 200
n_t = 1000
p = 20
sigma_e = 0.25
quantile_factor = np.arange(0.5, 1.0, 0.05)
X_train, y_train, X_test, y_test = simulation_setup(n_i=n_i,
n_o=n_o,
n_t=n_t,
p=p,
sigma_e=sigma_e)
# print("Simulation setup: inliers: {}; outliers: {}; dimensions: {}".format(n_i, n_o, n_t, p, sigma_e))
# def get_metrics_matrix(quantile_factor):
# metrics_matrix = np.empty((len(quantile_factor), 3, 4))
# for i, q in enumerate(quantile_factor):
# unit_matrix = np.zeros((3, 4))
# clf_ifb = IFB()
# clf_ifb.fit(X_train, y_train, quantile_factor=q)
# beta_ifb = clf_ifb.beta
#
# clf_lr = LogisticRegression(fit_intercept=False, solver='lbfgs')
# clf_lr.fit(X_train, y_train)
#
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import classification_report
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
import matplotlib
matplotlib.use('Qt5Agg')
plt.style.use('ggplot')
n_i = 10000
n_o = 1000
p = 20
data_train, data_test, beta = simulation_setup(n_i=n_i,
n_o=n_o,
p=p,
sigma_o=100,
sigma_e=0.1)
df_train = pd.DataFrame(data=data_train)
df_test = pd.DataFrame(data=data_test)
# df_train = pd.read_csv('data_train.csv')
df_train[p] = df_train[p].astype(int)
# store the numpy array representation of the dataframe
X = df_train.drop([p], axis=1).values
me = MEstimator()
# estimated location of inliers
loc_estimated = me.loc_estimator(X)
scale_estimated = me.scale_estimator(X)
X_out = X[~inlier_flag]
if y is None:
if return_outliers:
return X_in, X_out
else:
return X_in
else:
y_in, y_out = y[inlier_flag], y[~inlier_flag]
if return_outliers:
return X_in, y_in, X_out, y_out
else:
return X_in, y_in
def fit_transform(self, X, y=None, return_outliers=False, n_inliers=None):
self.fit(X, y)
return self.transform(X,
y,
return_outliers=return_outliers,
n_inliers=n_inliers)
if __name__ == '__main__':
data = simulation_setup(n_i=1000, n_o=800, p=8)[0]
X, y = data[:, :-1], data[:, -1]
preprocessor = Preprocessor()
# X_in, y_in, X_out, y_out = preprocessor.fit_transform(X, y, return_outliers=True)
# # print('number of X_in: %d, y_in: %d' % (len(X_in), len(y_in)))
# # print('number of X_out: %d, y_out: %d' % (len(X_out), len(y_out)))
X_in = preprocessor.fit_transform(X)
print(X_in.shape)