import utility
import numpy
r, w = utility.get_R()
for k in [10, 50, 100]:
U, V = utility.nmf(r, k, w)
uv = numpy.dot(U, V)
lse = 0
for i in range(len(r)):
for j in range(len(r[i])):
if w[i, j] == 1:
lse += (r[i, j] - uv[i, j])**2
print 'For %d latent terms, LSE: %f' % (k, lse)
falsePos = 0;
falseNeg = 0;
precisionArray = []
recallArray = []
threshold_ranges = [x/10.0 for x in range(0, 50, 5)]
#threshold_ranges = [3, 3.5]
kf = KFold(100000, 10, True)
for train, test in kf:
for threshold in threshold_ranges:
test_index += 1
local_error = 0
r, w, test_rows = utility.r_skiplist(test)
u, v = utility.nmf(r, K_VALUE, w)
uv = numpy.dot(u, v)
# UV here is the 90% trained set. Comparison next
for row in test_rows:
ui = row[0]-1
mi = row[1]-1
if(uv[ui, mi] >= threshold):
if(r[ui, mi] >= threshold):
truePos += 1
else: falsePos += 1
else:
if(r[ui, mi] >= threshold):
#if(uv[ui, mi] < threshold):
# # Test for empty rows.
# for row in w:
# sum = 0
# for element in row:
# sum += element
# if sum == 0:
# print sum
#
# print "Switching to columns now"
# for column in w.T:
# sum = 0
# for element in column:
# sum += element
# if sum == 0:
# print sum
u, v = utility.nmf(r, K_VALUE, w)
uv = numpy.dot(u, v)
# UV here is the 90% trained set. Comparison next
for row in test_rows:
ui = row[0] - 1
mi = row[1] - 1
local_error += numpy.abs(r[ui, mi] - uv[ui, mi])
test_error.append(local_error / 10000)
print test_error
print max(test_error)
print min(test_error)
