def test_predict():
(I, J, K) = (5, 3, 2)
R = numpy.array(
[[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12], [13, 14, 15]],
dtype=float)
M = numpy.ones((I, J))
K = 3
U = numpy.array([[125., 126.], [126., 126.], [126., 126.], [126., 126.],
[126., 126.]])
V = numpy.array([[84., 84.], [84., 84.], [84., 84.]])
M_test = numpy.array([[0, 0, 1], [0, 1, 0], [0, 0, 0], [1, 1, 0],
[0, 0,
0]]) #R->3,5,10,11, R_pred->21084,21168,21168,21168
MSE = (444408561. + 447872569. + 447660964. + 447618649) / 4.
R2 = 1. - (444408561. + 447872569. + 447660964. + 447618649) / (
4.25**2 + 2.25**2 + 2.75**2 + 3.75**2) #mean=7.25
Rp = 357. / (
math.sqrt(44.75) * math.sqrt(5292.)
) #mean=7.25,var=44.75, mean_pred=21147,var_pred=5292, corr=(-4.25*-63 + -2.25*21 + 2.75*21 + 3.75*21)
nmf = NMF(R, M, K)
nmf.U = U
nmf.V = V
performances = nmf.predict(M_test)
assert performances['MSE'] == MSE
assert performances['R^2'] == R2
assert performances['Rp'] == Rp
def test_run():
# Data generated from W = [[1,2],[3,4]], H = [[4,3],[2,1]]
R = [[8, 5], [20, 13]]
M = [[1, 1], [1, 0]]
K = 2
U = numpy.array([[10, 9], [8, 7]], dtype='f') #2x2
V = numpy.array([[6, 4], [5, 3]], dtype='f') #2x2
nmf = NMF(R, M, K)
# Check we get an Exception if W, H are undefined
with pytest.raises(AssertionError) as error:
nmf.run(0)
assert str(
error.value
) == "U and V have not been initialised - please run NMF.initialise() first."
# Then check for 1 iteration whether the updates work - heck just the first entry of U
nmf.U = U
nmf.V = V
nmf.run(1)
U_00 = 10 * (6 * 8 / 96.0 + 5 * 5 / 77.0) / (5.0 + 6.0) #0.74970484061
assert abs(U_00 - nmf.U[0][0]) < 0.000001
def test_predict():
(I,J,K) = (5,3,2)
R = numpy.array([[1,2,3],[4,5,6],[7,8,9],[10,11,12],[13,14,15]],dtype=float)
M = numpy.ones((I,J))
K = 3
U = numpy.array([[125.,126.],[126.,126.],[126.,126.],[126.,126.],[126.,126.]])
V = numpy.array([[84.,84.],[84.,84.],[84.,84.]])
M_test = numpy.array([[0,0,1],[0,1,0],[0,0,0],[1,1,0],[0,0,0]]) #R->3,5,10,11, R_pred->21084,21168,21168,21168
MSE = (444408561. + 447872569. + 447660964. + 447618649) / 4.
R2 = 1. - (444408561. + 447872569. + 447660964. + 447618649) / (4.25**2+2.25**2+2.75**2+3.75**2) #mean=7.25
Rp = 357. / ( math.sqrt(44.75) * math.sqrt(5292.) ) #mean=7.25,var=44.75, mean_pred=21147,var_pred=5292, corr=(-4.25*-63 + -2.25*21 + 2.75*21 + 3.75*21)
nmf = NMF(R,M,K)
nmf.U = U
nmf.V = V
performances = nmf.predict(M_test)
assert performances['MSE'] == MSE
assert performances['R^2'] == R2
assert performances['Rp'] == Rp
def test_compute_I_div():
R = [[1,2,0,4],[5,0,7,0]]
M = [[1,1,0,1],[1,0,1,0]]
K = 2
U = numpy.array([[1,2],[3,4]],dtype='f') #2x2
V = numpy.array([[5,7,9,11],[6,8,10,12]],dtype='f').T #4x2
#R_pred = [[17,23,29,35],[39,53,67,81]]
expected_I_div = sum([
1.0*math.log(1.0/17.0) - 1.0 + 17.0,
2.0*math.log(2.0/23.0) - 2.0 + 23.0,
4.0*math.log(4.0/35.0) - 4.0 + 35.0,
5.0*math.log(5.0/39.0) - 5.0 + 39.0,
7.0*math.log(7.0/67.0) - 7.0 + 67.0
])
nmf = NMF(R,M,K)
nmf.U = U
nmf.V = V
I_div = nmf.compute_I_div()
assert I_div == expected_I_div
def test_run():
# Data generated from W = [[1,2],[3,4]], H = [[4,3],[2,1]]
R = [[8,5],[20,13]]
M = [[1,1],[1,0]]
K = 2
U = numpy.array([[10,9],[8,7]],dtype='f') #2x2
V = numpy.array([[6,4],[5,3]],dtype='f') #2x2
nmf = NMF(R,M,K)
# Check we get an Exception if W, H are undefined
with pytest.raises(AssertionError) as error:
nmf.run(0)
assert str(error.value) == "U and V have not been initialised - please run NMF.initialise() first."
# Then check for 1 iteration whether the updates work - heck just the first entry of U
nmf.U = U
nmf.V = V
nmf.run(1)
U_00 = 10*(6*8/96.0+5*5/77.0)/(5.0+6.0) #0.74970484061
assert abs(U_00 - nmf.U[0][0]) < 0.000001
def test_compute_I_div():
R = [[1, 2, 0, 4], [5, 0, 7, 0]]
M = [[1, 1, 0, 1], [1, 0, 1, 0]]
K = 2
U = numpy.array([[1, 2], [3, 4]], dtype='f') #2x2
V = numpy.array([[5, 7, 9, 11], [6, 8, 10, 12]], dtype='f').T #4x2
#R_pred = [[17,23,29,35],[39,53,67,81]]
expected_I_div = sum([
1.0 * math.log(1.0 / 17.0) - 1.0 + 17.0,
2.0 * math.log(2.0 / 23.0) - 2.0 + 23.0,
4.0 * math.log(4.0 / 35.0) - 4.0 + 35.0,
5.0 * math.log(5.0 / 39.0) - 5.0 + 39.0,
7.0 * math.log(7.0 / 67.0) - 7.0 + 67.0
])
nmf = NMF(R, M, K)
nmf.U = U
nmf.V = V
I_div = nmf.compute_I_div()
assert I_div == expected_I_div
