def test_initialisation():
I,J = 2,3
R = numpy.ones((I,J))
M = numpy.ones((I,J))
K = 4
L = 5
# Init FG ones, S random
init_FG = 'ones'
init_S = 'random'
nmtf = NMTF(R,M,K,L)
nmtf.initialise(init_S,init_FG)
assert numpy.array_equal(numpy.ones((I,K)),nmtf.F)
assert numpy.array_equal(numpy.ones((J,L)),nmtf.G)
for (k,l) in itertools.product(range(0,K),range(0,L)):
assert nmtf.S[k,l] > 0 and nmtf.S[k,l] < 1
# Init FG random, S ones
init_FG = 'random'
init_S = 'ones'
nmtf = NMTF(R,M,K,L)
nmtf.initialise(init_S,init_FG)
assert numpy.array_equal(numpy.ones((K,L)),nmtf.S)
for (i,k) in itertools.product(range(0,I),range(0,K)):
assert nmtf.F[i,k] > 0 and nmtf.F[i,k] < 1
for (j,l) in itertools.product(range(0,J),range(0,L)):
assert nmtf.G[j,k] > 0 and nmtf.G[j,k] < 1
# Init FG kmeans, S exponential
init_FG = 'kmeans'
init_S = 'exponential'
nmtf = NMTF(R,M,K,L)
nmtf.initialise(init_S,init_FG)
for (i,k) in itertools.product(range(0,I),range(0,K)):
assert nmtf.F[i,k] == 0.2 or nmtf.F[i,k] == 1.2
for (j,l) in itertools.product(range(0,J),range(0,L)):
assert nmtf.G[j,k] == 0.2 or nmtf.G[j,k] == 1.2
for (k,l) in itertools.product(range(0,K),range(0,L)):
assert nmtf.S[k,l] > 0
# Load in data
R = numpy.loadtxt(input_folder + "R.txt")
M = numpy.ones((I, J))
# Run the VB algorithm, <repeats> times
times_repeats = []
performances_repeats = []
for i in range(0, repeats):
# Set all the seeds
numpy.random.seed(3)
random.seed(4)
scipy.random.seed(5)
# Run the classifier
nmtf = NMTF(R, M, K, L)
nmtf.initialise(init_S, init_FG, expo_prior)
nmtf.run(iterations)
# Extract the performances and timestamps across all iterations
times_repeats.append(nmtf.all_times)
performances_repeats.append(nmtf.all_performances)
# Check whether seed worked: all performances should be the same
assert all([numpy.array_equal(performances, performances_repeats[0]) for performances in performances_repeats]), \
"Seed went wrong - performances not the same across repeats!"
# Print out the performances, and the average times
all_times_average = list(numpy.average(times_repeats, axis=0))
all_performances = performances_repeats[0]
print "np_all_times_average = %s" % all_times_average
print "np_all_performances = %s" % all_performances
# We now run the VB algorithm on each of the M's for each fraction.
all_performances = {metric: [] for metric in metrics}
average_performances = {metric: []
for metric in metrics} # averaged over repeats
for (fraction, Ms, Ms_test) in zip(fractions_unknown, all_Ms, all_Ms_test):
print "Trying fraction %s." % fraction
# Run the algorithm <repeats> times and store all the performances
for metric in metrics:
all_performances[metric].append([])
for (repeat, M, M_test) in zip(range(0, repeats), Ms, Ms_test):
print "Repeat %s of fraction %s." % (repeat + 1, fraction)
# Run the VB algorithm
nmtf = NMTF(R, M, K, L)
nmtf.initialise(init_S, init_FG)
nmtf.run(iterations)
# Measure the performances
performances = nmtf.predict(M_test)
for metric in metrics:
# Add this metric's performance to the list of <repeat> performances for this fraction
all_performances[metric][-1].append(performances[metric])
# Compute the average across attempts
for metric in metrics:
average_performances[metric].append(
sum(all_performances[metric][-1]) / repeats)
print "repeats=%s \nfractions_unknown = %s \nall_performances = %s \naverage_performances = %s" % \