def do_run():
print('Random\n')
dis.pairwise(dat.matrix)
ndis.vecpair(dat.matrix)
cov.correl(dat.matrix)
ncov.veccor(dat.matrix)
entrop(dat.matrix)
''' print('\nIris\n')
def test_algorithms():
n = 10
delta = 0.1
pmodel = pairwise(n)
k = 1
# generated a BTL model with probabilites close to one and zero
pmodel.generate_deterministic_BTL([k * i / float(n) for i in range(n)])
print("largest entry: ", amax(pmodel.P))
print("model complexity: ", pmodel.top1H())
kset = [1, n]
ar = ARalg(pmodel, kset)
ar.rank(0.1)
print('AR, # Comparisons:', ar.pairwise.ctr)
print('..succeeded' if ar.evaluate_perfect_recovery() else '..failed')
plpac = PLPAC(pmodel)
plpac.rank(delta)
print('PLPAC, # Comparisons:', plpac.pairwise.ctr)
print('..succeeded' if plpac.evaluate_perfect_recovery() else '..failed')
alg = topkalg(pmodel, 1)
alg.rank()
print('AR2, #comparisons:', alg.pairwise.ctr)
print('..succeeded' if alg.evaluate_perfect_recovery() else '..failed')
btm = BTM(pmodel)
btm.rank(delta)
print('BTM, #comparisons:', btm.pairwise.ctr)
print('..succeeded' if btm.evaluate_perfect_recovery() else '..failed')
def reproduce_figure_selection_confidence_interval():
n = 5
pmodel = pairwise(n)
pmodel.generate_const(0.1)
k = 2
rule = 4
alg = topkalg(pmodel, k, rule)
varydelta(alg, "./dat/cmprules3")
def get_points(arrangement, table):
arrangement = list(arrangement)
arrangement.append(arrangement[0])
points = 0
for a, b in pairwise(arrangement):
points += get_points_for_a_next_to_b(a, b, table)
return points
def exp_fig_varyn(ns, const=5, nit=300, std=0):
k = 1
rule = 7
#ns = [10,15,20,25,30,35,40]
models = [pairwise(n) for n in ns]
for pmodel in models:
pmodel.generate_deterministic_BTL(
[const * i / float(pmodel.n**1.1) for i in range(pmodel.n)])
pmodel.uniform_perturb(std)
print(amax(pmodel.P))
print(pmodel.scores())
pmodel = pairwise(2)
alg = topkalg(pmodel, k, rule)
plpac = PLPAC(pmodel)
btm = BTM(pmodel)
algorithms = [alg, plpac, btm]
result = compare_models(models, algorithms, nit, False)
ar = result[:, 3]
plpac = result[:, 5]
btmb = result[:, 7]
return result
def exp_fig6():
#n = 5
n = 10
delta = 0.1
#nit = 500
nit = 10
#kset = range(0,n*(n-1)/2+1)
kset = range(0, 15)
result = zeros((len(kset), 12))
for k in kset:
sampcomp = [[], [], []]
successp = [[], [], []]
for it in range(nit):
# for each instance, generate a random model...
pmodel = pairwise(n)
pmodel.generate_deterministic_BTL([log(1 + i) for i in range(n)])
oncemore = True
while pmodel.top1H() > 250000 or oncemore:
oncemore = False
# find off diagonals
offdiags = []
for i in range(0, n):
offdiags += [(i, j) for j in range(i + 1, n)]
offdiags = random.permutation(offdiags)
for (i, j) in offdiags[:k]:
pmodel.P[i, j] = 0.5 * random.rand()
pmodel.P[j, i] = 1 - pmodel.P[i, j]
pmodel.sortP() # to make sure 0 is the top item
print(pmodel.top1H())
alg = topkalg(pmodel, 1)
plpac = PLPAC(pmodel, pmodel.top1H() * 50)
btm = BTM(pmodel, pmodel.top1H() * 50)
for nalg, alg in enumerate([alg, plpac, btm]):
alg.rank(delta)
sampcomp[nalg].append(alg.pairwise.ctr / pmodel.top1H())
successp[nalg].append(alg.evaluate_perfect_recovery())
print(nalg, alg.pairwise.ctr / pmodel.top1H())
print(
nalg, 'succeeded'
if alg.evaluate_perfect_recovery() else 'failed')
result[k, 0] = k / float(n * (n - 1) / 2)
for nalg in range(3):
result[k, 3 + 2 * nalg] = mean(sampcomp[nalg])
result[k, 3 + 2 * nalg + 1] = sqrt(var(sampcomp[nalg]))
result[k, 9 + nalg] = 1 - mean(successp[nalg])
savetxt("./fig/generalization_n10.dat", result, delimiter='\t')
def exp_revision(relative=False):
n = 10
#nit = 400
nit = 50
ks = range(1, 130, 10)
models = [pairwise(n) for i in ks]
for pmodel, k in zip(models, ks):
pmodel.generate_deterministic_BTL(
[log(0.09 * k + i) for i in range(n)])
alg = topkalg(pmodel, 1, 7) # original AR algorithm
savage = topkalg(pmodel, 1, 2) # SAVAGE algorithm from Urvoy et al. 2013
algorithms = [alg, savage]
result = compare_models(models, algorithms, nit, relative)
savetxt("./dat/comparison_vary_closeness_linsep_rev.dat",
result,
delimiter='\t')
def exp_fig4b(relative=False):
n = 10
nit = 200
kset = range(1, n)
models = [pairwise(n) for k in kset]
for pmodel, k in zip(models, kset):
pmodel.generate_deterministic_BTL(
[0.6 * k * i / float(n) for i in range(n)])
k = 1
rule = 7
alg = topkalg(pmodel, k, rule)
plpac = PLPAC(pmodel)
btm = BTM(pmodel)
algorithms = [alg, plpac, btm]
result = compare_models(models, algorithms, nit, relative)
savetxt("./fig/comparison_vary_closeness_extreme.dat",
result,
delimiter='\t')
def exp_fig4a(relative=False):
n = 10
#nit = 400
nit = 200
ks = range(1, 130, 10)
#ks = [1]
models = [pairwise(n) for i in ks]
for pmodel, k in zip(models, ks):
pmodel.generate_deterministic_BTL(
[log(0.09 * k + i) for i in range(n)])
k = 1
rule = 7
alg = topkalg(pmodel, k, rule)
plpac = PLPAC(pmodel)
btm = BTM(pmodel)
#alg2 = topkalg(pmodel,6)
#algorithms = [alg,plpac,btm,alg2]
algorithms = [alg, plpac, btm]
result = compare_models(models, algorithms, nit, relative)
savetxt("./fig/comparison_vary_closeness_linsep.dat",
result,
delimiter='\t')
import EntropyComputation as entr
import datums as dat
import pairwise as dis
import pairwiseVectorization as ndis
import covariance as cov
import CorrelationVectorized as ncov
print('Random\n')
dis.pairwise(dat.matrix)
ndis.vecpair(dat.matrix)
cov.correl(dat.matrix)
ncov.veccor(dat.matrix)
# print('\nIris\n')
# dis.pairwise(dat.iris)
# ndis.vecpair(dat.iris)
# cov.correl(dat.iris)
# ncov.veccor(dat.iris)
# print('\nBreast Cancer\n')
# dis.pairwise(dat.breast_cancer)
# ndis.vecpair(dat.breast_cancer)
# cov.correl(dat.breast_cancer)
# ncov.veccor(dat.breast_cancer)
# print('\nDigits\n')
# dis.pairwise(dat.digits)
# ndis.vecpair(dat.digits)
# cov.correl(dat.digits)
# ncov.veccor(dat.digits)
p = PorterStemmer()
index = {}
count_term = {}
start_time = time.time()
with open(stopwords_file, "r") as file:
stopwords = map(lambda line: line.strip(), file.readlines())
with open(index_file, "r") as index_file:
lines = index_file.readlines()
for line in lines:
entry = line.split(" ")
documents = entry[2:]
dictionary = defaultdict(int)
for document, count in pairwise(documents):
dictionary[document] = int(count)
count_term.update({entry[0]: int(entry[1])})
index.update({entry[0]: dictionary})
with open(lengths_file, "r") as lengths:
documents_lengths = map(lambda line: line.strip(), lengths.readlines())
documents_lengths = [int(length) for length in documents_lengths]
corpus_length = documents_lengths[len(documents_lengths) - 1]
del documents_lengths[len(documents_lengths) - 1]
for term in query:
term = term.translate(None, string.punctuation)
if term not in stopwords:
inner_query.append(p.stem(term.lower(), 0, len(term) - 1))
def entrop(X):
print('starting running .....')
np.random.seed(100)
params = range(10, 141, 10) # different param setting
nparams = len(params) # number of different parameters
perf_loop = np.zeros([10, nparams
]) # 10 trials = 10 rows, each parameter is a column
perf_cool = np.zeros([10, nparams])
counter = 0
for ncols in X:
nrows = len(X[0])
print("matrix dimensions: ", nrows, ncols)
for i in range(10):
#X = np.random.randint(0,20,[nrows,ncols]) # random matrix
# you need to use random.rand(...) for float matrix
st = time.time()
entropy_dist = dis.pairwise(X)
et = time.time()
perf_loop[i, counter] = et - st # time difference
st = time.time()
entropy_npdist = ndis.vecpair(X)
et = time.time()
perf_cool[i, counter] = et - st
assert np.isclose(entropy_loop, entropy_cool, atol=1e-06)
counter = counter + 1
mean_loop = np.mean(
perf_loop,
axis=0) # mean time for each parameter setting (over 10 trials)
mean_cool = np.mean(perf_cool, axis=0)
std_loop = np.std(perf_loop, axis=0) # standard deviation
std_cool = np.std(perf_cool, axis=0)
import matplotlib.pyplot as plt
plt.errorbar(params,
mean_loop[0:nparams],
yerr=std_loop[0:nparams],
color='red',
label='Loop Solution')
plt.errorbar(params,
mean_cool[0:nparams],
yerr=std_cool[0:nparams],
color='blue',
label='Matrix Solution')
plt.xlabel('Number of Cols of the Matrix')
plt.ylabel('Running Time (Seconds)')
plt.legend()
plt.savefig('CompareEntropyFig.pdf')
# plt.show() # uncomment this if you want to see it right way
print("result is written to CompareEntropyFig.pdf")
