def getpairs(self, batch, params):
g1 = []
g2 = []
for i in batch:
g1.append(i[0].embeddings)
g2.append(i[1].embeddings)
g1x, g1mask = self.prepare_data(g1)
g2x, g2mask = self.prepare_data(g2)
embg1 = self.feedforward_function(g1x, g1mask)
embg2 = self.feedforward_function(g2x, g2mask)
for idx, i in enumerate(batch):
i[0].representation = embg1[idx, :]
i[1].representation = embg2[idx, :]
pairs = utils.get_pairs_fast(batch, params.samplingtype)
p1 = []
p2 = []
for i in pairs:
p1.append(i[0].embeddings)
p2.append(i[1].embeddings)
p1x, p1mask = self.prepare_data(p1)
p2x, p2mask = self.prepare_data(p2)
return (g1x, g1mask, g2x, g2mask, p1x, p1mask, p2x, p2mask)
def get_pairs(self, batch, params):
g1 = []
g2 = []
g1_s = []
g2_s = []
for i in batch:
g1.append(i[0].embeddings)
g2.append(i[1].embeddings)
g1_s.append(i[0].phrase)
g2_s.append(i[1].phrase)
g1_ = [g1[i:i + params.batchsize] for i in range(0, len(g1), params.batchsize)]
g2_ = [g2[i:i + params.batchsize] for i in range(0, len(g2), params.batchsize)]
embg1 = []
embg2 = []
for i in range(len(g1_)):
g1x, g1mask = self.prepare_data(g1_[i])
g2x, g2mask = self.prepare_data(g2_[i])
embg1_ = self.feedforward_function(g1x, g1mask)
embg2_ = self.feedforward_function(g2x, g2mask)
embg1.append(embg1_)
embg2.append(embg2_)
embg1 = np.vstack(embg1)
embg2 = np.vstack(embg2)
#update representations
for idx, i in enumerate(batch):
i[0].representation = embg1[idx, :]
i[1].representation = embg2[idx, :]
pairs = utils.get_pairs_fast(batch, params.samplingtype)
p1 = []
p2 = []
p1_s = []
p2_s = []
for i in pairs:
p1.append(i[0].embeddings)
p2.append(i[1].embeddings)
p1_s.append(i[0].phrase)
p2_s.append(i[1].phrase)
p1x, p1mask = self.prepare_data(p1)
p2x, p2mask = self.prepare_data(p2)
g1x, g1mask = self.prepare_data(g1)
g2x, g2mask = self.prepare_data(g2)
return (g1x, g1mask, g2x, g2mask, p1x, p1mask, p2x, p2mask),(g1_s,g2_s,p1_s,p2_s)
def get_pairs(self, batch, params):
g1n = []
g1w = []
g1l = []
g2n = []
g2w = []
g2l = []
for i in batch:
if params.combination_type != "ngram-word-lstm":
g1n.append(i[0].embeddings)
g1w.append(i[1].embeddings)
g2n.append(i[2].embeddings)
g2w.append(i[3].embeddings)
else:
g1n.append(i[0].embeddings)
g1w.append(i[1].embeddings)
g1l.append(i[2].embeddings)
g2n.append(i[3].embeddings)
g2w.append(i[4].embeddings)
g2l.append(i[5].embeddings)
g1nx, g1nmask = self.prepare_data(g1n)
g1wx, g1wmask = self.prepare_data(g1w)
g2nx, g2nmask = self.prepare_data(g2n)
g2wx, g2wmask = self.prepare_data(g2w)
if params.combination_type == "ngram-word-lstm":
g1lx, g1lmask = self.prepare_data(g1l)
g2lx, g2lmask = self.prepare_data(g2l)
if params.combination_type != "ngram-word-lstm":
embg1 = self.feedforward_function(g1nx, g1nmask, g1wx, g1wmask)
embg2 = self.feedforward_function(g2nx, g2nmask, g2wx, g2wmask)
else:
embg1 = self.feedforward_function(g1nx, g1nmask, g1wx, g1wmask,
g1lx, g1lmask)
embg2 = self.feedforward_function(g2nx, g2nmask, g2wx, g2wmask,
g2lx, g2lmask)
if params.combination_type != "ngram-word-lstm":
batch_n = []
batch_w = []
for idx, i in enumerate(batch):
i[0].representation = embg1[idx, :]
i[1].representation = embg1[idx, :]
i[2].representation = embg2[idx, :]
i[3].representation = embg2[idx, :]
batch_n.append((i[0], i[2]))
batch_w.append((i[1], i[3]))
else:
batch_n = []
batch_w = []
batch_l = []
for idx, i in enumerate(batch):
i[0].representation = embg1[idx, :]
i[1].representation = embg1[idx, :]
i[2].representation = embg1[idx, :]
i[3].representation = embg2[idx, :]
i[4].representation = embg2[idx, :]
i[5].representation = embg2[idx, :]
batch_n.append((i[0], i[3]))
batch_w.append((i[1], i[4]))
batch_l.append((i[2], i[5]))
pairs1 = utils.get_pairs_fast(batch_n, params.samplingtype)
p1n = []
p2n = []
for i in pairs1:
p1n.append(i[0].embeddings)
p2n.append(i[1].embeddings)
p1nx, p1nmask = self.prepare_data(p1n)
p2nx, p2nmask = self.prepare_data(p2n)
pairs2 = utils.get_pairs_fast(batch_w, params.samplingtype)
p1w = []
p2w = []
for i in pairs2:
p1w.append(i[0].embeddings)
p2w.append(i[1].embeddings)
p1wx, p1wmask = self.prepare_data(p1w)
p2wx, p2wmask = self.prepare_data(p2w)
if params.combination_type == "ngram-word-lstm":
pairs3 = utils.get_pairs_fast(batch_l, params.samplingtype)
p1l = []
p2l = []
for i in pairs3:
p1l.append(i[0].embeddings)
p2l.append(i[1].embeddings)
p1lx, p1lmask = self.prepare_data(p1l)
p2lx, p2lmask = self.prepare_data(p2l)
if params.combination_type != "ngram-word-lstm":
return (g1nx, g1nmask, g1wx, g1wmask, g2nx, g2nmask, g2wx, g2wmask,
p1nx, p1nmask, p1wx, p1wmask, p2nx, p2nmask, p2wx, p2wmask)
else:
return (g1nx, g1nmask, g1wx, g1wmask, g1lx, g1lmask, g2nx, g2nmask,
g2wx, g2wmask, g2lx, g2lmask, p1nx, p1nmask, p1wx, p1wmask,
p1lx, p1lmask, p2nx, p2nmask, p2wx, p2wmask, p2lx, p2lmask)