def one_run_m():
print "start materialized regression"
m_regressor = NormalizedLogisticRegression()
start = time.time()
m_regressor.fit(T, Y, w_init=w_init)
end = time.time()
print "end materialized regression"
return end - start
def one_run_n():
print "start factorized regression"
n_regressor = NormalizedLogisticRegression()
start = time.time()
n_regressor.fit(normalized_matrix, Y, w_init=w_init2)
end = time.time()
print "end factorized regression"
return end - start
print T.shape w_init = np.matrix(np.random.randn(T.shape[1], 1)) w_init2 = np.matrix(w_init, copy=True) gamma = 0.000001 iterations = 20 result_eps = 1e-6 print "start factorized matrix" normalized_matrix = nm.NormalizedMatrix(s, [r1, r2], k) print "end factorized matrix" import time print "start materialized regression" m_regressor = NormalizedLogisticRegression() start = time.time() m_regressor.fit(T, Y, w_init=w_init) end = time.time() print "end materialized regression" m_time = end - start print "start factorized regression" n_regressor = NormalizedLogisticRegression() start = time.time() n_regressor.fit(normalized_matrix, Y, w_init=w_init2) end = time.time() print "end factorized regression" n_time = end - start print "speedup is", m_time / n_time
dr = ds * f
ns = nr * t
s = np.random.rand(ns, ds)
r = [np.random.rand(nr, dr)]
num = np.random.randint(nr, size=ns)
while (max(num) != nr - 1):
num = np.random.randint(nr, size=ns)
k = [num]
T = np.mat(np.hstack((s, r[0][k[0]])))
Y = np.matrix(np.random.randint(2, size=ns)).T
normalized_matrix = nm.NormalizedMatrix(s, r, k)
avg = []
for _ in range(trails):
w_init_m = np.matrix(np.random.randn(T.shape[1], 1))
m_regressor = NormalizedLogisticRegression()
m_start = time.time()
m_regressor.fit(T, Y, w_init=w_init_m)
m_end = time.time()
w_init_n = np.matrix(np.random.randn(T.shape[1], 1))
n_regressor = NormalizedLogisticRegression()
n_start = time.time()
n_regressor.fit(normalized_matrix, Y, w_init=w_init_n)
n_end = time.time()
avg.append((m_end - m_start) / (n_end - n_start))
print(sum(avg) - min(avg) - max(avg)) / (trails - 2)
result.append((sum(avg) - min(avg) - max(avg)) / (trails - 2))
total.append(result)