def argMax_XleftOvers(left, s, wd, wr, query):
"""grid search over ranked list to find most
diverse and relevant doc at a time"""
if len(s) == 0:
return 0, 10.0 # choose the first one
else:
f_measures = []
# compute f_measure to all left docs
for i, doc in enumerate(left):
hs_ofR = compute_hsofR(doc, s, query)
doc_r = doc_it(doc)
f_measure = np.dot(wr, doc_r) + np.dot(wd, hs_ofR)
f_measures.append([i, f_measure])
# rank by highest f_measure
f_measures.sort(key=operator.itemgetter(1), reverse=True)
ind, gain = f_measures[0]
#print ind, gain, str(left[ind]["id"])
return ind, gain
def argMax_XleftOvers(left, s, wd, wr, query):
"""grid search over ranked list to find most
diverse and relevant doc at a time"""
if len(s) == 0:
return 0, 10.0 # choose the first one
else:
f_measures = []
# compute f_measure to all left docs
for i, doc in enumerate(left):
hs_ofR = compute_hsofR(doc, s, query)
doc_r = doc_it(doc)
f_measure = np.dot(wr, doc_r) + np.dot(wd, hs_ofR)
f_measures.append([i, f_measure])
# rank by highest f_measure
f_measures.sort(key=operator.itemgetter(1), reverse=True)
ind, gain = f_measures[0]
#print ind, gain, str(left[ind]["id"])
return ind, gain
def argMax_XleftOvers(left, s, wd, wr, query):
"""grid search over ranked list to find most
diverse and relevant doc at a time"""
if len(s) == 0:
return 0, 10.0 # choose the first one
else:
f_measures = []
# compute f_measure to all left docs
for i, doc in enumerate(left):
doc = {u'nbComments': u'0', u'cn': [u'0.11766567164179105', u'0.0022447761194029849', u'0.062770149253731339', u'0.1366686567164179', u'0.056316417910447759', u'1.7910447761194031e-05', u'0.012674626865671642', u'0.00020298507462686568', u'0.00011940298507462687', u'0.60114029850746264', u'0.010179104477611941'], u'date_taken': u'2005-01-31T14:49:41Z', u'faceProportion': u'0.029029029029029', u'rank': u'143', u'poi': u'agra fort', u'id': u'50729250', u'tagSpecificity': u'0.581247562023133', u'title': u'Agra Fort', u'tags': 'travel india agra unesco worldheritage uttarpradesh', u'location': u'0,0', u'meanImageTagClarity': u'26.8629387557', u'username': u'Hyougushi', u'description': u'\u3059\u3079\u3066\u5927\u7406\u77f3\u3002 (20050129 Agra, Uttar Pradesh, INDIA) * UNESCO World Heritage * Agra Fort (1983) \u4e16\u754c\u907a\u7523\u300c\u30a2\u30fc\u30b0\u30e9\u57ce\u585e\u300d', u'views': u'0', u'locationSimilarity': u'0', u'meanTagRank': u'8062.62546455', u'uniqueTags': u'0.012318630168325283', u'license': u'5', u'userid': u'71481771@N00', u'meanPhotoViews': u'84.8545', u'visualScore': u'0.683108042839217', u'url_b': u'http://farm1.static.flickr.com/25/50729250_5a039d278e_b.jpg'}
hs_ofR = compute_hsofR(doc, s, query)
doc_r = doc_it(doc)
f_measure = np.dot(wr, doc_r) + np.dot(wd, hs_ofR)
f_measures.append([i, f_measure])
# rank by highest f_measure
f_measures.sort(key=operator.itemgetter(1), reverse=True)
ind, gain = f_measures[0]
#print ind, gain, str(left[ind]["id"])
return ind, gain
def argMax_XleftOvers(left, s, wd, wr, query):
"""grid search over ranked list to find most
diverse and relevant doc at a time"""
if len(s) == 0:
return 0, 10.0 # choose the first one
else:
f_measures = []
# compute f_measure to all left docs
for i, doc in enumerate(left):
doc = {
u'nbComments':
u'0',
u'cn': [
u'0.11766567164179105', u'0.0022447761194029849',
u'0.062770149253731339', u'0.1366686567164179',
u'0.056316417910447759', u'1.7910447761194031e-05',
u'0.012674626865671642', u'0.00020298507462686568',
u'0.00011940298507462687', u'0.60114029850746264',
u'0.010179104477611941'
],
u'date_taken':
u'2005-01-31T14:49:41Z',
u'faceProportion':
u'0.029029029029029',
u'rank':
u'143',
u'poi':
u'agra fort',
u'id':
u'50729250',
u'tagSpecificity':
u'0.581247562023133',
u'title':
u'Agra Fort',
u'tags':
'travel india agra unesco worldheritage uttarpradesh',
u'location':
u'0,0',
u'meanImageTagClarity':
u'26.8629387557',
u'username':
u'Hyougushi',
u'description':
u'\u3059\u3079\u3066\u5927\u7406\u77f3\u3002 (20050129 Agra, Uttar Pradesh, INDIA) * UNESCO World Heritage * Agra Fort (1983) \u4e16\u754c\u907a\u7523\u300c\u30a2\u30fc\u30b0\u30e9\u57ce\u585e\u300d',
u'views':
u'0',
u'locationSimilarity':
u'0',
u'meanTagRank':
u'8062.62546455',
u'uniqueTags':
u'0.012318630168325283',
u'license':
u'5',
u'userid':
u'71481771@N00',
u'meanPhotoViews':
u'84.8545',
u'visualScore':
u'0.683108042839217',
u'url_b':
u'http://farm1.static.flickr.com/25/50729250_5a039d278e_b.jpg'
}
hs_ofR = compute_hsofR(doc, s, query)
doc_r = doc_it(doc)
f_measure = np.dot(wr, doc_r) + np.dot(wd, hs_ofR)
f_measures.append([i, f_measure])
# rank by highest f_measure
f_measures.sort(key=operator.itemgetter(1), reverse=True)
ind, gain = f_measures[0]
#print ind, gain, str(left[ind]["id"])
return ind, gain
wd_q = np.zeros(3)
wr_q = np.zeros(wr_len) # TBD how to represent
doc_set = X[query] # given a ordered doc set
s = []
for i, doc in enumerate(doc_set):
if not s: # if s is empty of docs
s.append(doc)
continue
else:
# compute wd_q for one
hs_ofR = compute_hsofR(doc, s, query)
# compute wr_q for one
doc_r = doc_it(doc)
# compute wd_q for rest
wd_rest = compute_wd(doc_set[i:], s, query, wd_q, wr_q)
# compute wr_q for rest
wr_rest = compute_wr(doc_set[i:], s, query, wd_q, wr_q)
# compute denominator
denom = compute_denom(doc_set[i:], s, query, wd_q, wr_q)
# subtract wd_q
wd_q += wd_rest / denom - hs_ofR
# subtract wr_q
wr_q += wr_rest / denom - doc_r
# update w's
s.append(doc)
print query