def word_distance(w1, w2, wv):
"""
# word_distance: spectral word distance
#
# Input:
# w1 - a word (string)
# w2 - a word (string)
# wv - a dictionary with keys words and values embedding vectors
#
# Output:
# dist - a floating point number in the range (0, 1)
#
# Instructions:
# You should use the cosine function (imported at the beginning of this file as
# cs), to compute cosine angle similarity between the two words.
"""
dist = 0.0
# -----
# YOUR CODE GOES HERE
v1 = wv[w1]
v2 = wv[w2]
dist = cs(v1, v2)
# -----
return dist
def calc_distances(dic1,dic2,task,topks,distances,GT_samples):
logger = logging.getLogger("my_logger")
logger.info('TASK: %s',task)
maps =[]
for dis in distances:
logger.info('distance: %s',dis)
tops_val = []
for topk in topks:
logger.info('K: %s',topk)
mAP = 0
order_of_samples1 = sorted(dic1.keys())
order_of_samples2 = sorted(dic2.keys())
for sample in order_of_samples1:
score_samples = []
reps1 = dic1[sample]
for given_sample in order_of_samples2:
if sample == given_sample:
pass
else:
reps2 = dic2[given_sample]
if dis == 'cos':
given_score = 1 - cs(reps1, reps2)
elif dis == 'sp':
given_score = sp.entropy(reps1, reps2)
elif dis == 'corr':
given_score = pearsonr(reps1,reps2)[0] #1-->correlated, 0-->no correlated
else:
given_score = eu(reps1, reps2)
score_samples.append((given_sample, given_score))
if dis == 'corr':
sorted_scores = sorted(score_samples, key=lambda x:x[1],reverse=True) #decreasing
else:
sorted_scores = sorted(score_samples, key=lambda x:x[1],reverse=False) #increasing
mAP = mAP + get_AP(sorted_scores, sample, GT_samples, top_k=topk)
tops_val.append(mAP/(len(dic1.keys())*1.0))
logger.info('mAP %s %s',task, str(tops_val[-1]))
maps.append(tops_val)
maps = pd.DataFrame(maps)
maps.columns = topks
maps['distances'] = distances
maps['task'] = task
return maps