ee364a_hw4_prob5p13.py

cens_fit_data.py

import numpy as np
import numpy.matlib as ml

# data for censored fitting problem.
np.random.seed(0)

n = 20  # dimension of x's
M = 25  # number of non-censored data points
K = 100 # total number of points
c_true = ml.randn(n, 1)
X_unsorted = ml.randn(n, K)
y_unsorted = X_unsorted.T * c_true + 0.1 * np.sqrt(n) * ml.randn(K, 1)

# Reorder measurements, then censor
sort_ind = np.argsort(y_unsorted, 0)
y = y_unsorted[sort_ind, 0]
X = X_unsorted[:, sort_ind]
D = 0.5 * (y[M - 1] + y[M])
y = y[0:M]

ee364a_hw4_prob5p13.py

# In[34]:

from cvxpy import *
from cens_fit_data import *
#from numpy.linalg import norm


# In[35]:

c = Variable(n)
y_cens = Variable(K - M)
constraints = [y_cens >= D]


# In[36]:

uncensored_error = sum(square(y - X[:, 0:M].T * c))
censored_error = sum(square(y_cens - X[:, M::].T * c))
objective = Minimize(uncensored_error + censored_error)

problem_with_censored = Problem(objective, constraints)
problem_with_censored.solve()
#print('c Error when including censored data: {}'.format(np.linalg.norm(c_true - c.value) / np.linalg.norm(c_true)))


# In[28]:

error = norm2(vstack(y, y_cens) - X.T * c)
objective = Minimize(error)

problem_with_censored = Problem(objective, constraints)
problem_with_censored.solve(solver=CVXOPT)


# In[36]:

uncensored_error = (y - X[:, 0:M].T * c).T * (y - X[:, 0:M].T * c)
censored_error = (y_cens - X[:, M::].T * c).T * (y_cens - X[:, M::].T * c)
objective = Minimize(uncensored_error + censored_error)

problem_with_censored = Problem(objective, constraints)
problem_with_censored.solve()


# In[ ]: