Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Sign up
Appearance settings

Fix TabErrors in three files #5

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
lihanghang merged 4 commits into lihanghang:master from cclauss:patch-1
Aug 1, 2019
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
56 changes: 26 additions & 30 deletions Algorithm-code/RelationExtraction_CNN+ATT/main_att.py
View file Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,52 +10,51 @@


def collate_fn(batch):

data, label = zip(*batch)
return data, label

def test(**kwargs):
pass

def train(**kwargs):
kwargs.update({'model': 'PCNN_ATT'})
opt.parse(kwargs)
if opt.use_gpu:
kwargs.update({'model': 'PCNN_ATT'})
opt.parse(kwargs)
if opt.use_gpu:
torch.cuda.set_device(opt.gpu_id)

model = getattr(models, 'PCNN_ATT')(opt)
if opt.use_gpu:
model = getattr(models, 'PCNN_ATT')(opt)
if opt.use_gpu:
model.cuda()

# loading data
DataModel = getattr(datasets, opt.data + 'Data')
train_data = DataModel(opt.data_root, train=True)
train_data_loader = DataLoader(train_data, opt.batch_size, shuffle=True, num_workers=opt.num_workers, collate_fn = collate_fn)
test_data = DataModel(opt.data_root, train=False)
test_data_loader = DataLoader(test_data, opt.batch_size, shuffle=False, num_workers=opt.num_workers, collate_fn = collate_fn)
print('{} train data: {}; test data: {}'.format(now(), len(train_data), len(test_data)))
DataModel = getattr(datasets, opt.data + 'Data')
train_data = DataModel(opt.data_root, train=True)
train_data_loader = DataLoader(train_data, opt.batch_size, shuffle=True, num_workers=opt.num_workers, collate_fn = collate_fn)

test_data = DataModel(opt.data_root, train=False)
test_data_loader = DataLoader(test_data, opt.batch_size, shuffle=False, num_workers=opt.num_workers, collate_fn = collate_fn)
print('{} train data: {}; test data: {}'.format(now(), len(train_data), len(test_data)))

optimizer = optim.Adadelta(model.parameters(), rho=0.95, eps=1e-6)
optimizer = optim.Adadelta(model.parameters(), rho=0.95, eps=1e-6)


# train
for epoch in range(opt.num_epochs):
# train
for epoch in range(opt.num_epochs):
total_loss = 0
for idx, (data, label_set) in enumerate(train_data_loader):

label = [l[0] for l in label_set]
label = [l[0] for l in label_set]

optimizer.zero_grad()
model.batch_size = opt.batch_size
loss = model(data, label)
if opt.use_gpu:
label = torch.LongTensor(label).cuda()
else:
label = torch.LongTensor(label)
loss.backward()
optimizer.step()
total_loss += loss.item()
optimizer.zero_grad()
model.batch_size = opt.batch_size
loss = model(data, label)
if opt.use_gpu:
label = torch.LongTensor(label).cuda()
else:
label = torch.LongTensor(label)
loss.backward()
optimizer.step()
total_loss += loss.item()

if epoch > 2:
pred_res, p_num = predict_var(model, test_data_loader)
Expand Down Expand Up @@ -145,6 +144,3 @@ def predict(model, test_data_loader):
if __name__ == '__main__':
import fire
fire.Fire()



162 changes: 76 additions & 86 deletions Algorithm-code/RelationExtraction_CNN+ATT/models/PCNN_ATT.py
View file Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,15 +8,12 @@
from torch.autograd import Variable




'''
References:
《Neural Relation Extraction with Selective Attention over Instances》--PCNN+ATT(2016)
《Distant Supervision for Relation Extraction via Piecewise Convolutional Neural Networks》--PCNN(2015)
'''
class PCNN_ATT(BasicModule):


def __init__(self, opt):
super(PCNN_ATT, self).__init__()
Expand Down Expand Up @@ -62,45 +59,43 @@ def __init__(self, opt):
def init_model_weight(self):
"""
here use xavier approach
"""
"""
nn.init.xavier_uniform(self.rel_embs)
nn.init.uniform(self.rel_bias)
for conv in self.convs:
nn.init.xavier_uniform(conv.weight)
nn.init.uniform(conv.bias)


nn.init.xavier_uniform(conv.weight)
nn.init.uniform(conv.bias)

def init_word_emb(self):
def p_2norm(path):
v = torch.from_numpy(np.load(path))
if self.opt.norm_emb:
v = torch.div(v, v.norm(2 ,1).unsqueeze(1))
v[v != v] == 0.0
v = torch.div(v, v.norm(2 ,1).unsqueeze(1))
v[v != v] == 0.0
return v

w2v = p_2norm(self.opt.w2v_path)
p1_2v = p_2norm(self.opt.p1_2v_path)
p2_2v = p_2norm(self.opt.p2_2v_path)

if self.opt.use_gpu:
self.word_embs.weight.data.copy_(w2v.cuda())
self.pos1_embs.weight.data.copy_(p1_2v.cuda())
self.pos2_embs.weight.data.copy_(p2_2v.cuda())
self.word_embs.weight.data.copy_(w2v.cuda())
self.pos1_embs.weight.data.copy_(p1_2v.cuda())
self.pos2_embs.weight.data.copy_(p2_2v.cuda())
else:
self.pos1_embs.weight.data.copy_(p1_2v)
self.pos2_embs.weight.data.copy_(p2_2v)
self.word_embs.weight.data.copy_(w2v)
self.pos1_embs.weight.data.copy_(p1_2v)
self.pos2_embs.weight.data.copy_(p2_2v)
self.word_embs.weight.data.copy_(w2v)


def init_int_constant(self, num):
'''
a util function for generating a LongTensor Variable
'''
if self.opt.use_gpu:
return Variable(torch.LongTensor([num]).cuda())
else:
return Variable(torch.LongTensor([num]))
def init_int_constant(self, num):
'''
a util function for generating a LongTensor Variable
'''
if self.opt.use_gpu:
return Variable(torch.LongTensor([num]).cuda())
else:
return Variable(torch.LongTensor([num]))


def mask_piece_pooling(self, x, mask):
Expand All @@ -125,110 +120,105 @@ def piece_max_pooling(self, x, insPool):
batch_res = []

for i in range(split_pool):
ins = split_batch_x[i].squeeze()
pool = split_pool[i].squeeze().data
seg_1 = ins[:, :pool[0]].max(1)[0].unsqueeze(1)
seg_2 = ins[:, pool[0]: pool[1]][0].unsqueeze(1)
seg_3 = ins[:, pool[1]:].max(1)[0].unsqueeze(1)
piece_max_pool = torch.cat([seg_1, seg_2, seg_3], 1).view(1, -1)
batch_res.append(piece_max_pool)
ins = split_batch_x[i].squeeze()
pool = split_pool[i].squeeze().data
seg_1 = ins[:, :pool[0]].max(1)[0].unsqueeze(1)
seg_2 = ins[:, pool[0]: pool[1]][0].unsqueeze(1)
seg_3 = ins[:, pool[1]:].max(1)[0].unsqueeze(1)
piece_max_pool = torch.cat([seg_1, seg_2, seg_3], 1).view(1, -1)
batch_res.append(piece_max_pool)

out = torch.cat(batch_res, 0)
assert out.size(1) == 3 * self.opt.filters_num
return out



def forward(self, x, label=None):
# get all sentences embedding in all bags of one batch
self.bags_feature = self.get_bags_features(x)
# get all sentences embedding in all bags of one batch
self.bags_feature = self.get_bags_features(x)

if label is None:
# for test
assert self.training is False
return self.test(x)
else:
# for train
assert self.training is True
return self.fit(x, label)
if label is None:
# for test
assert self.training is False
return self.test(x)
else:
# for train
assert self.training is True
return self.fit(x, label)


def fit(self, x, label):
'''
train process
'''
'''
train process
'''
x = self.get_batch_feature(label)
x = self.dropout(x)
out = x.mm(self.rel_embs.t()) + self.rel_bias
if self.opt.use_gpu:
v_label = torch.LongTensor(label).cuda()
v_label = torch.LongTensor(label).cuda()
else:
v_label = torch.LongTensor(label)
v_label = torch.LongTensor(label)
ce_loss = F.cross_entropy(out, Variable(v_label))
return ce_loss


def test(self, x):
'''
'''
test process
'''
pre_y = []
for label in range(0, self.opt.rel_num):
lables = [lables for _ in range(len(x))]
bags_feature = self.get_bathch_feature(lables)
out = self.test_scale_p * bags_feature.mm(self.rel_embs.t()) + self.rel_bias
pre_y.append(out.unsqueeze(1))
'''
pre_y = []
for label in range(0, self.opt.rel_num):
lables = [label for _ in range(len(x))]
bags_feature = self.get_bathch_feature(lables)
out = self.test_scale_p * bags_feature.mm(self.rel_embs.t()) + self.rel_bias
pre_y.append(out.unsqueeze(1))
res = torch.cat(pre_y, 1).max(1)[0]
return F.softmax(res, 1).t()


def get_batch_feature(self, lables):
'''
'''
Using Attention to get all bags embedding in a batch
'''
batch_feature = []
for bag_embs, label in zip(self.bags_feature, lables):
alpha = bag_embs.mm(self.rel_embs[label].view(-1, 1))
bag_embs = bag_embs * F.softmax(alpha, 0)
bag_vec = torch.sum(bag_embs, 0)
batch_feature.append(bag_vec.unsqueeze(0))
'''
batch_feature = []
for bag_embs, label in zip(self.bags_feature, lables):
alpha = bag_embs.mm(self.rel_embs[label].view(-1, 1))
bag_embs = bag_embs * F.softmax(alpha, 0)
bag_vec = torch.sum(bag_embs, 0)
batch_feature.append(bag_vec.unsqueeze(0))

return torch.cat(batch_feature, 0)
return torch.cat(batch_feature, 0)


def get_bags_features(self, bags):
'''
get all bags embedding in one batch before Attention
'''
get all bags embedding in one batch before Attention
'''
bags_feature = []
for bag in bags:
if self.opt.use_gpu:
data = map(lambda x: Variable(torch.LongTensor(x).cuda()), bag)
else:
data = map(lambda x: Variable(torch.LongTensor(x)), bag)
data = map(lambda x: Variable(torch.LongTensor(x)), bag)
bag_embs = self.get_ins_emb(data)
bags_feature.append(bag_embs)

return bags_feature


def get_ins_emb(self, x):
'''
'''
x: all instance in a Bag
'''
insEnt, _, insX, insPFs, insPool, mask = x
insPF1, insPF2 = [i.squeeze(1) for i in torch.split(insPFs, 1, 1)]

word_emb = self.word_embs(insX)
pf1_emb = self.pos1_embs(insPF1)
pf2_emb = self.pos2_embs(insPF2)

x = torch.cat([word_emb, pf1_emb, pf2_emb], 2)
x = x.unsqueeze(1)
x = [conv(x).squeeze(3) for conv in self.convs]
x = [self.mask_piece_pooling(i, mask) for i in x]
x = torch.cat(x, 1).tanh()
return x



'''
insEnt, _, insX, insPFs, insPool, mask = x
insPF1, insPF2 = [i.squeeze(1) for i in torch.split(insPFs, 1, 1)]

word_emb = self.word_embs(insX)
pf1_emb = self.pos1_embs(insPF1)
pf2_emb = self.pos2_embs(insPF2)

x = torch.cat([word_emb, pf1_emb, pf2_emb], 2)
x = x.unsqueeze(1)
x = [conv(x).squeeze(3) for conv in self.convs]
x = [self.mask_piece_pooling(i, mask) for i in x]
x = torch.cat(x, 1).tanh()
return x
12 changes: 6 additions & 6 deletions Algorithm-code/RelationExtraction_CNN+ATT/utils.py
View file Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@ def save_pr(out_dir, name, epoch, pre, rec, fp_res=None, opt=None):
fp_out = open('{}/{}_{}_FP.txt'.format(out_dir, name, epoch + 1), 'w')
for idx, r, p in fp_res:
fp_out.write('{}{}{}\n'.format(idx, r, p))
fp_out.close()
fp_out.close()

for p, r in zip(pre, rec):
out.write('{} {}\n'.format(p, r))
Expand Down Expand Up @@ -61,11 +61,11 @@ def eval_metric(true_y, pred_y, pred_p):
precision = 1.0
else:
precision = tp * 1.0 / (fp + tp)
recall = tp * 1.0 / positive_num
if precision != all_pre[-1] or recall != all_rec[-1]:
all_pre.append(precision)
all_rec.append(recall)
recall = tp * 1.0 / positive_num
if precision != all_pre[-1] or recall != all_rec[-1]:
all_pre.append(precision)
all_rec.append(recall)


print("tp={}; fp={}; fn={}; positive_num={}".format(tp, fp, fn, positive_num))
return all_pre[1:], all_rec[1:], fp_res
return all_pre[1:], all_rec[1:], fp_res

AltStyle によって変換されたページ (->オリジナル) /