def wordsimi353(embeddingDict, average_emb, rand, dim, ML):
#embeddingDict=loadEmbeddingFile('/mounts/data/proj/wenpeng/Emb_Extend/results_L2norm_forall/conc_0_1_2_3_4.txt')
#embeddingDict=loadEmbeddingFile('/mounts/data/proj/wenpeng/Tensor/result/tensor_ppmi/word2embedding_giga_wiki_20141201_300d.txt')
#embeddingDict=loadEmbeddingFile(filename)
#emb_matrix=mat(embeddingDict.values())
#norm_vector=LA.norm(emb_matrix, ord=2,axis=0)
#emb_normalized=numpy.divide(emb_matrix,norm_vector)
#key_list=embeddingDict.keys()
wordPairs2label = load_wordsimi353()
wordPair2simi = {}
labels = []
predicts = []
unknown = 0
for (word1, word2), label in wordPairs2label.iteritems():
#labels.append(label)
embedding1 = embeddingDict.get(word1)
embedding2 = embeddingDict.get(word2)
if embedding1 is not None and embedding2 is not None:
predict = dot_prod(embedding1, embedding2)
wordPair2simi[(word1, word2)] = predict
predicts.append(predict)
labels.append(label)
else:
if not ML:
if embedding1 is None:
if rand:
vector = numpy.random.uniform(-1, 1, dim)
embedding1 = list(vector / numpy.linalg.norm(vector))
else:
embedding1 = average_emb
if embedding2 is None:
if rand:
vector = numpy.random.uniform(-1, 1, dim)
embedding2 = list(vector / numpy.linalg.norm(vector))
else:
embedding2 = average_emb
predict = dot_prod(embedding1, embedding2)
wordPair2simi[(word1, word2)] = predict
predicts.append(predict)
labels.append(label)
unknown += 1
sp, p_value = spearmanr(labels, predicts)
print 'Spearmanr is: ' + str(sp) + ', unknown pairs: ' + str(unknown)
def wordsimi353(embeddingDict,average_emb,rand,dim,ML):
#embeddingDict=loadEmbeddingFile('/mounts/data/proj/wenpeng/Emb_Extend/results_L2norm_forall/conc_0_1_2_3_4.txt')
#embeddingDict=loadEmbeddingFile('/mounts/data/proj/wenpeng/Tensor/result/tensor_ppmi/word2embedding_giga_wiki_20141201_300d.txt')
#embeddingDict=loadEmbeddingFile(filename)
#emb_matrix=mat(embeddingDict.values())
#norm_vector=LA.norm(emb_matrix, ord=2,axis=0)
#emb_normalized=numpy.divide(emb_matrix,norm_vector)
#key_list=embeddingDict.keys()
wordPairs2label=load_wordsimi353()
wordPair2simi={}
labels=[]
predicts=[]
unknown=0
for (word1, word2), label in wordPairs2label.iteritems():
#labels.append(label)
embedding1=embeddingDict.get(word1)
embedding2=embeddingDict.get(word2)
if embedding1 is not None and embedding2 is not None:
predict=dot_prod(embedding1, embedding2)
wordPair2simi[(word1, word2)]=predict
predicts.append(predict)
labels.append(label)
else:
if not ML:
if embedding1 is None:
if rand:
vector=numpy.random.uniform(-1,1,dim)
embedding1=list(vector / numpy.linalg.norm(vector))
else:
embedding1=average_emb
if embedding2 is None:
if rand:
vector=numpy.random.uniform(-1,1,dim)
embedding2=list(vector / numpy.linalg.norm(vector))
else:
embedding2=average_emb
predict=dot_prod(embedding1, embedding2)
wordPair2simi[(word1, word2)]=predict
predicts.append(predict)
labels.append(label)
unknown+=1
sp,p_value=spearmanr(labels, predicts)
print 'Spearmanr is: '+str(sp)+', unknown pairs: '+str(unknown)
def RW(embeddingDict, average_emb, rand, dim, ML):
#embeddingDict=loadEmbeddingFile(filename)
#embeddingDict=loadEmbeddingFile('/mounts/data/proj/wenpeng/Tensor/result/tensor_ppmi/word2embedding_giga_wiki_20141201_300d.txt')
#embeddingDict=loadEmbeddingFile('/mounts/data/proj/wenpeng/Dataset/glove.42B.300d.txt') #hlbl-embeddings-original.EMBEDDING_SIZE=100.txt')
wordPairs2label = load_RW()
wordPair2simi = {}
labels = []
predicts = []
unknown = 0
for (word1, word2), label in wordPairs2label.iteritems():
#labels.append(label)
embedding1 = embeddingDict.get(word1)
embedding2 = embeddingDict.get(word2)
if embedding1 is not None and embedding2 is not None:
predict = dot_prod(embedding1, embedding2)
wordPair2simi[(word1, word2)] = predict
predicts.append(predict)
labels.append(label)
else:
if not ML:
if embedding1 is None:
if rand:
vector = numpy.random.uniform(-1, 1, dim)
embedding1 = list(vector / numpy.linalg.norm(vector))
else:
embedding1 = average_emb
if embedding2 is None:
if rand:
vector = numpy.random.uniform(-1, 1, dim)
embedding2 = list(vector / numpy.linalg.norm(vector))
else:
embedding2 = average_emb
predict = dot_prod(embedding1, embedding2)
wordPair2simi[(word1, word2)] = predict
predicts.append(predict)
labels.append(label)
unknown += 1
sp, p_value = spearmanr(labels, predicts)
print 'Spearmanr is: ' + str(sp) + ', unknown pairs: ' + str(unknown)
def RW(embeddingDict,average_emb,rand,dim,ML):
#embeddingDict=loadEmbeddingFile(filename)
#embeddingDict=loadEmbeddingFile('/mounts/data/proj/wenpeng/Tensor/result/tensor_ppmi/word2embedding_giga_wiki_20141201_300d.txt')
#embeddingDict=loadEmbeddingFile('/mounts/data/proj/wenpeng/Dataset/glove.42B.300d.txt') #hlbl-embeddings-original.EMBEDDING_SIZE=100.txt')
wordPairs2label=load_RW()
wordPair2simi={}
labels=[]
predicts=[]
unknown=0
for (word1, word2), label in wordPairs2label.iteritems():
#labels.append(label)
embedding1=embeddingDict.get(word1)
embedding2=embeddingDict.get(word2)
if embedding1 is not None and embedding2 is not None:
predict=dot_prod(embedding1, embedding2)
wordPair2simi[(word1, word2)]=predict
predicts.append(predict)
labels.append(label)
else:
if not ML:
if embedding1 is None:
if rand:
vector=numpy.random.uniform(-1,1,dim)
embedding1=list(vector / numpy.linalg.norm(vector))
else:
embedding1=average_emb
if embedding2 is None:
if rand:
vector=numpy.random.uniform(-1,1,dim)
embedding2=list(vector / numpy.linalg.norm(vector))
else:
embedding2=average_emb
predict=dot_prod(embedding1, embedding2)
wordPair2simi[(word1, word2)]=predict
predicts.append(predict)
labels.append(label)
unknown+=1
sp,p_value=spearmanr(labels, predicts)
print 'Spearmanr is: '+str(sp)+', unknown pairs: '+str(unknown)