def matrix_completion(self):
if self.completion == 'matrix':
res, _ = md.matrix_decomposition(self.Y, self.mask, **self.options)
elif self.completion == 'mean':
means = self.Y.mean(0) # means along the column axis
# calculate the true means along the column axis
# using only the values in the mask
rows, cols = self.Y.shape
means = sp.zeros(cols)
# calculate the true means along the column axis
# using only the values in the mask
for c in range(cols):
sum = 0.0
n = 0
for r in range(rows):
if self.mask[r][c]:
sum += self.Y[r][c]
n += 1
if n == 0:
means[c] = 0
else:
means[c] = sum/n
res = double(self.Y)
for r in range(rows):
for c in range(cols):
if not self.mask[r][c]:
res[r][c] = means[c]
return res
def comparative_test(self):
TH, GA = self._TH, self._GA
Y, Mask, args = self._Y, self._Mask, self._decomposition_args
A, B = md.matrix_decomposition(Y, Mask, **args)
if self.verbose == True:
print GA[1, :]
print B[1, :]
print Mask[1, :]
print TH[1, :]
print A[1, :]
error = ((TH - A)**2).sum() / TH.size
print "mean square error:", error
error2 = ((TH - Y*Mask)**2).sum() / TH.size
print "mse for naive solution:", error2
print "improvement:", error2/error, "times"
def test_decomposition_id(self):
# testing using the identity map
TH = sp.array([[1, 2, 3], [2, 4, 6]])
GA = sp.array([[0, 0, 0], [0, 100, 0]])
Y = TH + GA
A, B = md.matrix_decomposition(Y, lambda_d=0.1, mu_d=0.08, alpha=1000)
self.assertLess(abs(TH-A).sum(), 1.0, "")
self.assertLess(abs(TH-A).max(), 0.2, "")
self.assertLess(abs(GA-A).sum() < 1.0, "")
def matrix_completion(self):
if self.completion == 'matrix':
res, _ = md.matrix_decomposition(self.Y, self.mask, **self.options)
elif self.completion == 'mean':
means = self.Y.mean(0) # means along the column axis
# calculate the true means along the column axis
# using only the values in the mask
rows, cols = self.Y.shape
means = sp.zeros(cols)
# calculate the true means along the column axis
# using only the values in the mask
for c in range(cols):
sum = 0.0
n = 0
for r in range(rows):
if self.mask[r][c]:
sum += self.Y[r][c]
n += 1
if n == 0:
means[c] = 0
else:
means[c] = sum / n
res = double(self.Y)
for r in range(rows):
for c in range(cols):
if not self.mask[r][c]:
res[r][c] = means[c]
return res
def test_decomposition_id(self):
# testing using the identity map
TH = sp.array([[1, 2, 3], [2, 4, 6]])
GA = sp.array([[0, 0, 0], [0, 100, 0]])
Y = TH + GA
A, B = md.matrix_decomposition(Y, lambda_d=0.1, mu_d=0.08, alpha=1000)
self.assertLess(abs(TH - A).sum(), 1.0, "")
self.assertLess(abs(TH - A).max(), 0.2, "")
self.assertLess(abs(GA - A).sum() < 1.0, "")
def comparative_test(self):
TH, GA = self._TH, self._GA
Y, Mask, args = self._Y, self._Mask, self._decomposition_args
A, B = md.matrix_decomposition(Y, Mask, **args)
if self.verbose == True:
print GA[1, :]
print B[1, :]
print Mask[1, :]
print TH[1, :]
print A[1, :]
error = ((TH - A)**2).sum() / TH.size
print "mean square error:", error
error2 = ((TH - Y * Mask)**2).sum() / TH.size
print "mse for naive solution:", error2
print "improvement:", error2 / error, "times"
