def test_different_penalty(self):
'''Tests different penalty values'''
l1 = [1, 2, 3, 4]
l2 = [2, 5, 4, 3]
dktau_p1 = topk.kendall_tau_distance(l1, l2, p=1)
dktau_p0 = topk.kendall_tau_distance(l1, l2, p=0)
dktau_p05 = topk.kendall_tau_distance(l1, l2, p=0.5)
self.assertTrue(dktau_p1 != dktau_p0)
self.assertTrue(dktau_p1 != dktau_p05)
self.assertTrue(dktau_p0 != dktau_p05)
def test_different_k(self):
'''Tests different k values'''
l1 = [1, 2, 3, 4]
l2 = [2, 5, 4, 3]
dktau_k2 = topk.kendall_tau_distance(l1, l2, k=2)
dktau_k3 = topk.kendall_tau_distance(l1, l2, k=3)
dktau_kneg1 = topk.kendall_tau_distance(l1, l2, k=-1)
self.assertTrue(dktau_k2 != dktau_k3)
self.assertTrue(dktau_k2 != dktau_kneg1)
self.assertTrue(dktau_k3 != dktau_kneg1)
def test_different_penalty(self):
'''Tests different penalty values'''
l1 = [1, 2, 3, 4]
l2 = [2, 5, 4, 3]
dktau_p1 = topk.kendall_tau_distance(l1, l2, p=1)
dktau_p0 = topk.kendall_tau_distance(l1, l2, p=0)
dktau_p05 = topk.kendall_tau_distance(l1, l2, p=0.5)
self.assertTrue(dktau_p1 != dktau_p0)
self.assertTrue(dktau_p1 != dktau_p05)
self.assertTrue(dktau_p0 != dktau_p05)
def test_different_k(self):
'''Tests different k values'''
l1 = [1, 2, 3, 4]
l2 = [2, 5, 4, 3]
dktau_k2 = topk.kendall_tau_distance(l1, l2, k=2)
dktau_k3 = topk.kendall_tau_distance(l1, l2, k=3)
dktau_kneg1 = topk.kendall_tau_distance(l1, l2, k=-1)
self.assertTrue(dktau_k2 != dktau_k3)
self.assertTrue(dktau_k2 != dktau_kneg1)
self.assertTrue(dktau_k3 != dktau_kneg1)
def main(args=None):
if not args: args = []
if len(args) != 2:
print('Usage %s <in file>' % sys.argv[0])
return 1
in_fpath = args[1]
#Each row will have 10 cols:
#user, tag, value, baseline, mean_tag_prob, tag_freq, \
#item_tag_prob, item_tag_prob, tag_item_freq, user_tags
data = np.genfromtxt(in_fpath, skip_footer=1)
users = np.unique(data[:,0])
to_plot_corr = []
for i, user in enumerate(users):
#select based on first column
udata = data[data[:,0] == user]
our_method = udata[:,2]
baseline = 1.0 / udata[:,3]
#If the first is nan, user has too many tags leading to probs = 0
if our_method[0] == np.nan:
continue
rank_our = udata[our_method.argsort(),:][:,1]
rank_base = udata[baseline.argsort(),:][:,1]
print(i, rank_our[:10])
print(i, rank_base[:10])
corr = topk.kendall_tau_distance(rank_our, rank_base)
to_plot_corr.append(corr)
corr_data = np.array(to_plot_corr)
mean = np.mean(corr_data)
std = np.std(corr_data)
skew = stats.skew(corr_data)
_10perc = stats.scoreatpercentile(corr_data, 10)
_50perc = stats.scoreatpercentile(corr_data, 50)
_90perc = stats.scoreatpercentile(corr_data, 90)
print('Mean %.3f' % mean)
print('Std %.3f' % std)
print('Skew %.3f' % skew)
print('10%% Percentile %.3f' % _10perc)
print('50%% Percentile %.3f' % _50perc)
print('90%% Percentile %.3f' % _90perc)
print()
def test_all_paper_values(self):
'''
Tests based on the datasets by the paper:
[1] Algorithms for Estimating Relative Importance in Networks.
Scott White, Padhraic Smyth
KDD 2003
Values will not match the paper exactly because we use a different
normalization.
'''
Khemais = 1
Beghal = 2
Moussaoui = 3
Maaroufi = 4
Qatada = 5
Daoudi = 6
Courtaillier = 7
Bensakhria = 8
Walid = 9
Khammoun = 10
Atta = 11
Al_Shehhi = 12
al_Shibh = 13
Jarrah = 14
Hanjour = 15
Al_Omari = 16
Bahaji = 17
list_page_rank = [
Khemais, Beghal, Moussaoui, Maaroufi, Qatada, Daoudi, Courtaillier,
Bensakhria, Walid, Khammoun
]
list_wkpaths = [
Beghal, Khemais, Moussaoui, Maaroufi, Bensakhria, Daoudi, Qatada,
Walid, Courtaillier, Khammoun
]
list_markov_c = [
Atta, Al_Shehhi, al_Shibh, Moussaoui, Jarrah, Hanjour, Al_Omari,
Khemais, Qatada, Bahaji
]
self.assertEquals(
0, topk.kendall_tau_distance(list_page_rank, list_page_rank))
self.assertEquals(
0, topk.kendall_tau_distance(list_wkpaths, list_wkpaths))
self.assertEquals(
0, topk.kendall_tau_distance(list_markov_c, list_markov_c))
self.assertEquals(
0.45,
topk.kendall_tau_distance(list_page_rank, list_page_rank[::-1]))
self.assertEquals(
0.45, topk.kendall_tau_distance(list_wkpaths, list_wkpaths[::-1]))
self.assertEquals(
0.45, topk.kendall_tau_distance(list_markov_c,
list_markov_c[::-1]))
self.assertAlmostEquals(
0.06, topk.kendall_tau_distance(list_page_rank, list_wkpaths), 2)
self.assertAlmostEquals(
0.68, topk.kendall_tau_distance(list_page_rank, list_markov_c), 2)
self.assertAlmostEquals(
0.06, topk.kendall_tau_distance(list_wkpaths, list_page_rank), 2)
self.assertAlmostEquals(
0.71, topk.kendall_tau_distance(list_wkpaths, list_markov_c), 2)
self.assertAlmostEquals(
0.68, topk.kendall_tau_distance(list_markov_c, list_page_rank), 2)
self.assertAlmostEquals(
0.71, topk.kendall_tau_distance(list_markov_c, list_wkpaths), 2)
self.assertEquals(
1, topk.kendall_tau_distance(list_page_rank, range(20, 30)))
self.assertEquals(
1, topk.kendall_tau_distance(list_wkpaths, range(20, 30)))
self.assertEquals(
1, topk.kendall_tau_distance(list_markov_c, range(20, 30)))
def test_all_paper_values(self):
'''
Tests based on the datasets by the paper:
[1] Algorithms for Estimating Relative Importance in Networks.
Scott White, Padhraic Smyth
KDD 2003
Values will not match the paper exactly because we use a different
normalization.
'''
Khemais = 1
Beghal = 2
Moussaoui = 3
Maaroufi = 4
Qatada = 5
Daoudi = 6
Courtaillier = 7
Bensakhria = 8
Walid = 9
Khammoun = 10
Atta = 11
Al_Shehhi = 12
al_Shibh = 13
Jarrah = 14
Hanjour = 15
Al_Omari = 16
Bahaji = 17
list_page_rank = [
Khemais,
Beghal,
Moussaoui,
Maaroufi,
Qatada,
Daoudi,
Courtaillier,
Bensakhria,
Walid,
Khammoun
]
list_wkpaths = [
Beghal,
Khemais,
Moussaoui,
Maaroufi,
Bensakhria,
Daoudi,
Qatada,
Walid,
Courtaillier,
Khammoun
]
list_markov_c = [
Atta,
Al_Shehhi,
al_Shibh,
Moussaoui,
Jarrah,
Hanjour,
Al_Omari,
Khemais,
Qatada,
Bahaji
]
self.assertEquals(0,
topk.kendall_tau_distance(list_page_rank, list_page_rank))
self.assertEquals(0,
topk.kendall_tau_distance(list_wkpaths, list_wkpaths))
self.assertEquals(0,
topk.kendall_tau_distance(list_markov_c, list_markov_c))
self.assertEquals(0.45,
topk.kendall_tau_distance(list_page_rank, list_page_rank[::-1]))
self.assertEquals(0.45,
topk.kendall_tau_distance(list_wkpaths, list_wkpaths[::-1]))
self.assertEquals(0.45,
topk.kendall_tau_distance(list_markov_c, list_markov_c[::-1]))
self.assertAlmostEquals(0.06,
topk.kendall_tau_distance(list_page_rank, list_wkpaths), 2)
self.assertAlmostEquals(0.68,
topk.kendall_tau_distance(list_page_rank, list_markov_c), 2)
self.assertAlmostEquals(0.06,
topk.kendall_tau_distance(list_wkpaths, list_page_rank), 2)
self.assertAlmostEquals(0.71,
topk.kendall_tau_distance(list_wkpaths, list_markov_c), 2)
self.assertAlmostEquals(0.68,
topk.kendall_tau_distance(list_markov_c, list_page_rank), 2)
self.assertAlmostEquals(0.71,
topk.kendall_tau_distance(list_markov_c, list_wkpaths), 2)
self.assertEquals(1,
topk.kendall_tau_distance(list_page_rank, range(20, 30)))
self.assertEquals(1,
topk.kendall_tau_distance(list_wkpaths, range(20, 30)))
self.assertEquals(1,
topk.kendall_tau_distance(list_markov_c, range(20, 30)))
