def batch_act(self, observations):
batchsize = len(observations)
# initialize a table of replies with this agent's id
batch_reply = [{'id': self.getID()} for _ in range(batchsize)]
# convert the observations into batches of inputs and targets
# `labels` stores the true labels returned in the `ys` vector
# `valid_inds` tells us the indices of all valid examples
# e.g. for input [{}, {'text': 'hello'}, {}, {}], valid_inds is [1]
# since the other three elements had no 'text' field
xs, ys, labels, valid_inds, is_training = self.vectorize(observations)
if xs is None:
# no valid examples, just return empty responses
return batch_reply
predictions = self.predict(xs, ys, is_training)
# maps returns predictions back to the right `valid_inds`
# in the example above, a prediction `world` should reply to `hello`
PaddingUtils.map_predictions(predictions.cpu().data,
valid_inds,
batch_reply,
observations,
self.dict,
self.END_IDX,
labels=labels,
answers=labels,
ys=ys.data if ys is not None else None,
report_freq=self.opt.get(
'report_freq', 0))
return batch_reply
def batch_act(self, observations):
batchsize = len(observations)
# initialize a table of replies with this agent's id
batch_reply = [{'id': self.getID()} for _ in range(batchsize)]
# convert the observations into batches of inputs and targets
# valid_inds tells us the indices of all valid examples
# e.g. for input [{}, {'text': 'hello'}, {}, {}], valid_inds is [1]
# since the other three elements had no 'text' field
xs, ys, labels, valid_inds, is_training = self.vectorize(observations)
if xs is None:
# no valid examples, just return empty responses
return batch_reply
# produce predictions, train on targets if availables
predictions = self.predict(xs, ys, is_training)
if is_training:
report_freq = 0
else:
report_freq = 0.01
PaddingUtils.map_predictions(predictions,
valid_inds,
batch_reply,
observations,
self.dict,
self.END_IDX,
report_freq=report_freq,
labels=labels,
answers=self.answers,
ys=ys.data)
return batch_reply
def batch_act(self, observations):
batch_reply = [{'id': self.getID()} for _ in range(len(observations))]
if any(['labels' in obs for obs in observations]):
# if we are starting a new training epoch, reinitialize hidden
if not self.is_training:
self.hidden = self.model.init_hidden(self.batchsize)
self.is_training = True
data_list, targets_list, _c, _v, y_lens = self.vectorize(
observations, self.opt['seq_len'], self.is_training)
else:
# if we just finished training, reinitialize hidden
if self.is_training:
self.hidden = self.model.init_hidden(self.batchsize)
self.is_training = False
data_list, targets_list, labels, valid_inds, y_lens = self.vectorize(
observations, self.opt['seq_len'], self.is_training)
if data_list is None:
# not enough data to batch act yet, return empty responses
return batch_reply
batch_reply = []
# during evaluation, len(data_list) is always 1
# during training, len(dat_list) >= 0: vectorize returns a list
# containing all batches available at the time it is called
for i in range(len(data_list)):
temp_dicts = [{
'id': self.getID()
} for _ in range(len(observations))]
# ignore case when we do not return any valid indices
if data_list[i] is not None:
output, hidden, predictions = self.predict(
data_list[i], self.hidden, targets_list[i],
self.is_training, y_lens)
self.hidden = self.repackage_hidden(hidden)
if predictions is not None:
# map predictions back to the right order
PaddingUtils.map_predictions(
predictions.cpu(),
valid_inds,
temp_dicts,
observations,
self.dict,
self.END_IDX,
report_freq=self.opt['report_freq'],
)
batch_reply += temp_dicts
# for prediction metrics computations, we get rid of PERSON1 and PERSON2 tokens
if not self.is_training:
for reply in batch_reply:
if 'text' in reply:
reply['text'] = reply['text'].replace('PERSON1 ', '')
reply['text'] = reply['text'].replace('PERSON2 ', '')
return batch_reply
def batch_act(self, observations):
batchsize = len(observations)
self.init_cuda_buffer(batchsize)
# initialize a table of replies with this agent's id
batch_reply = [{'id': self.getID()} for _ in range(batchsize)]
# convert the observations into batches of inputs and targets
# valid_inds tells us the indices of all valid examples
# e.g. for input [{}, {'text': 'hello'}, {}, {}], valid_inds is [1]
# since the other three elements had no 'text' field
print(observations)
xs, ys, labels, valid_inds, cands, valid_cands, is_training = self.vectorize(
observations)
print(xs)
if xs is None:
# no valid examples, just return empty responses
return batch_reply
# produce predictions, train on targets if availables
cand_inds = [i[0]
for i in valid_cands] if valid_cands is not None else None
predictions, cand_preds = self.predict(xs, ys, cands, cand_inds,
is_training)
if is_training:
report_freq = 0
else:
report_freq = self.report_freq
if predictions is not None:
PaddingUtils.map_predictions(
predictions,
valid_inds,
batch_reply,
observations,
self.dict,
self.END_IDX,
report_freq=report_freq,
labels=labels,
answers=self.answers,
ys=ys.data if ys is not None else None)
if cand_preds is not None:
if valid_cands is None:
valid_cands = [(None, i, labels) for i in valid_inds]
for i in range(len(valid_cands)):
order = cand_preds[i]
_, batch_idx, curr_cands = valid_cands[i]
curr = batch_reply[batch_idx]
curr['text_candidates'] = [
curr_cands[idx] for idx in order if idx < len(curr_cands)
]
return batch_reply
def vectorize(self, observations):
"""Convert a list of observations into input & target tensors."""
is_training = any(['labels' in obs for obs in observations])
xs, ys, labels, valid_inds, _, _ = PaddingUtils.pad_text(
observations,
self.dict,
end_idx=self.END_IDX,
null_idx=self.NULL_IDX,
dq=True,
eval_labels=True,
truncate=self.truncate)
if xs is None:
return None, None, None, None, None, None, None
xs = torch.LongTensor(xs)
if ys is not None:
ys = torch.LongTensor(ys)
if self.use_cuda:
# copy to gpu
self.xs.resize_(xs.size())
self.xs.copy_(xs)
xs = Variable(self.xs)
if ys is not None:
self.ys.resize_(ys.size())
self.ys.copy_(ys)
ys = Variable(self.ys)
else:
xs = Variable(xs)
if ys is not None:
ys = Variable(ys)
cands = None
valid_cands = None
if not is_training and self.rank:
# set up candidates
cands = []
valid_cands = []
for i, v in enumerate(valid_inds):
if 'label_candidates' in observations[v]:
curr_lcs = list(observations[v]['label_candidates'])
curr_cands = [{'text': c} for c in curr_lcs]
cs, _, _, valid_c_inds, *_ = PaddingUtils.pad_text(
curr_cands,
self.dict,
null_idx=self.NULL_IDX,
dq=True,
truncate=self.truncate)
valid_cands.append(
(i, v, [curr_lcs[j] for j in valid_c_inds]))
cs = torch.LongTensor(cs)
if self.use_cuda:
cs = cs.cuda()
cands.append(cs)
return xs, ys, labels, valid_inds, cands, valid_cands, is_training
def batch_act(self, observations):
batch_reply = [{'id': self.getID()} for _ in range(len(observations))]
if any(['labels' in obs for obs in observations]):
# if we are starting a new training epoch, reinitialize hidden
if self.is_training == False:
self.hidden = self.model.init_hidden(self.batchsize)
self.is_training = True
data_list, targets_list, _, _, y_lens = self.vectorize(
observations, self.opt['seq_len'], self.is_training)
else:
# if we just finished training, reinitialize hidden
if self.is_training == True:
self.hidden = self.model.init_hidden(self.batchsize)
self.is_training = False
data_list, targets_list, labels, valid_inds, y_lens = self.vectorize(
observations, self.opt['seq_len'], self.is_training)
if data_list is None:
# not enough data to batch act yet, return empty responses
return batch_reply
batch_reply = []
# during evaluation, len(data_list) is always 1
# during training, len(dat_list) >= 0: vectorize returns a list containing all batches available at the time it is called
for i in range(len(data_list)):
temp_dicts = [{
'id': self.getID()
} for _ in range(len(observations))]
output, hidden, loss_dict, predictions = self.predict(
data_list[i], self.hidden, targets_list[i], self.is_training,
y_lens)
self.hidden = self.repackage_hidden(hidden)
if predictions is not None:
# map predictions back to the right order
PaddingUtils.map_predictions(
predictions.cpu(),
valid_inds,
temp_dicts,
observations,
self.dict,
self.END_IDX,
report_freq=self.opt['report_freq'])
if loss_dict is not None:
if 'metrics' in temp_dicts[0]:
for k, v in loss_dict.items():
temp_dicts[0]['metrics'][k] = v
else:
temp_dicts[0]['metrics'] = loss_dict
batch_reply += temp_dicts
return batch_reply
def batch_act(self, observations):
batchsize = len(observations)
# initialize a table of replies with this agent's id
batch_reply = [{'id': self.getID()} for _ in range(batchsize)]
# convert the observations into batches of inputs and targets
# valid_inds tells us the indices of all valid examples
# e.g. for input [{}, {'text': 'hello'}, {}, {}], valid_inds is [1]
# since the other three elements had no 'text' field
xs, ys, labels, valid_inds, cands, valid_cands = self.vectorize(
observations)
if xs is None:
# no valid examples, just return empty responses
return batch_reply
# produce predictions, train on targets if availables
predictions, text_cand_inds, loss = self.predict(
xs, ys, cands, valid_cands)
if loss is not None:
if 'metrics' in batch_reply[0]:
for k, v in loss.items():
batch_reply[0]['metrics'][k] = v
else:
batch_reply[0]['metrics'] = loss
if ys is not None:
report_freq = 0
else:
report_freq = 0.1
PaddingUtils.map_predictions(predictions,
valid_inds,
batch_reply,
observations,
self.dict,
self.END_IDX,
report_freq=report_freq,
labels=labels,
answers=self.answers,
ys=ys)
if text_cand_inds is not None:
text_cand_inds = text_cand_inds.cpu().data
for i in range(len(valid_cands)):
order = text_cand_inds[i]
_, batch_idx, curr_cands = valid_cands[i]
curr = batch_reply[batch_idx]
curr['text_candidates'] = [
curr_cands[idx] for idx in order if idx < len(curr_cands)
]
return batch_reply
def batch_act(self, observations):
batchsize = len(observations)
# initialize a table of replies with this agent's id
batch_reply = [{'id': self.getID()} for _ in range(batchsize)]
# convert the observations into batches of inputs and targets
# valid_inds tells us the indices of all valid examples
# e.g. for input [{}, {'text': 'hello'}, {}, {}], valid_inds is [1]
# since the other three elements had no 'text' field
xs, ys, labels, valid_inds, cands, valid_cands, is_training = self.vectorize(observations)
if xs is None:
# no valid examples, just return empty responses
return batch_reply
# produce predictions, train on targets if availables
predictions, text_cand_inds = self.predict(xs, ys, cands, valid_cands, is_training)
self.distinct_ngrams(predictions)
if is_training:
report_freq = 0
else:
report_freq = self.report_freq
PaddingUtils.map_predictions(
predictions.cpu().data, valid_inds, batch_reply, observations,
self.dict, self.END_IDX, report_freq=report_freq, labels=labels,
answers=self.answers, ys=ys.data if ys is not None else None)
# tensor to list
list_ref = ys.cpu().data.numpy().tolist()
list_hypo = predictions.cpu().data.numpy().tolist()
# cut off from self.END_IDX
for ref in list_ref:
ind = self.end_idx(ref)
self.refs.append([ref[:ind]])
for hypo in list_hypo:
ind = self.end_idx(hypo)
self.hypos.append(hypo[:ind])
if text_cand_inds is not None:
text_cand_inds = text_cand_inds.cpu().data
for i in range(len(valid_cands)):
order = text_cand_inds[i]
_, batch_idx, curr_cands = valid_cands[i]
curr = batch_reply[batch_idx]
curr['text_candidates'] = [curr_cands[idx] for idx in order
if idx < len(curr_cands)]
return batch_reply
def vectorize(self, observations):
"""Convert a list of observations into input & target tensors."""
is_training = any(['labels' in obs for obs in observations])
xs, ys, labels, valid_inds, _, _ = PaddingUtils.pad_text(
observations,
self.dict,
end_idx=None,
null_idx=self.NULL_IDX,
dq=True,
eval_labels=True,
truncate=self.truncate)
if xs is None:
return None, None, None, None, None, None, None
xs = torch.LongTensor(xs)
ys = torch.LongTensor(ys)
if self.use_cuda:
# copy to gpu
self.xs.resize_(xs.size())
self.xs.copy_(xs)
xs = Variable(self.xs)
if ys is not None:
self.ys.resize_(ys.size())
self.ys.copy_(ys)
ys = Variable(self.ys)
else:
xs = Variable(xs)
if ys is not None:
ys = Variable(ys)
return xs, ys, labels, valid_inds, is_training
def vectorize(self, observations):
"""Convert a list of observations into input & target tensors."""
is_training = any(('labels' in obs for obs in observations))
# utility function for padding text and returning lists of indices
# parsed using the provided dictionary
xs, ys, labels, valid_inds, _, _ = PaddingUtils.pad_text(
observations,
self.dict,
end_idx=self.END_IDX,
null_idx=self.NULL_IDX,
dq=False,
eval_labels=True)
if xs is None:
return None, None, None, None, None
# move lists of indices returned above into tensors
xs = torch.LongTensor(xs)
if self.use_cuda:
xs = xs.cuda()
xs = Variable(xs)
if ys is not None:
ys = torch.LongTensor(ys)
if self.use_cuda:
ys = ys.cuda()
ys = Variable(ys)
return xs, ys, labels, valid_inds, is_training
def vectorize(self, observations):
"""Convert a list of observations into input & target tensors."""
ys = None
xs, ys, labels, valid_inds, _, _ = PaddingUtils.pad_text(
observations, self.dict, self.END_IDX, self.NULL_IDX, dq=True,
eval_labels=False, truncate=self.truncate)
if xs is None:
return None, None, None, None, None, None
if self.use_cuda:
# copy to gpu
self.xs.resize_(xs.size())
self.xs.copy_(xs, async=True)
xs = Variable(self.xs)
if ys is not None:
self.ys.resize_(ys.size())
self.ys.copy_(ys, async=True)
ys = Variable(self.ys)
else:
xs = Variable(xs)
if ys is not None:
ys = Variable(ys)
# set up candidates
cands = None
valid_cands = None
if ys is None and self.rank:
# only do ranking when no targets available and ranking flag set
parsed_cs = []
valid_cands = []
for i, v in enumerate(valid_inds):
if 'label_candidates' in observations[v]:
# each candidate tuple is a pair of the parsed version and
# the original full string
cs = list(observations[v]['label_candidates'])
curr_dqs = [deque(maxlen=self.truncate) for _ in cs]
for dq, c in zip(curr_dqs, cs):
dq.extendleft(reversed(self.parse(c)))
parsed_cs.append(curr_dqs)
valid_cands.append((i, v, cs))
if len(parsed_cs) > 0:
# TODO: store lengths of cands separately, so don't have zero
# padding for varying number of cands per example
# found cands, pack them into tensor
max_c_len = max(max(len(c) for c in cs) for cs in parsed_cs)
max_c_cnt = max(len(cs) for cs in parsed_cs)
for cs in parsed_cs:
for c in cs:
c += [self.NULL_IDX] * (max_c_len - len(c))
cs += [self.NULL_IDX] * (max_c_cnt - len(cs))
cands = torch.LongTensor(parsed_cs)
if self.use_cuda:
# copy to gpu
self.cands.resize_(cands.size())
self.cands.copy_(cands, async=True)
cands = Variable(self.cands)
else:
cands = Variable(cands)
return xs, ys, labels, valid_inds, cands, valid_cands
def batch_act(self, observations):
batchsize = len(observations)
# initialize a table of replies with this agent's id
batch_reply = [{'id': self.getID()} for _ in range(batchsize)]
# convert the observations into batches of inputs and targets
# valid_inds tells us the indices of all valid examples
# e.g. for input [{}, {'text': 'hello'}, {}, {}], valid_inds is [1]
# since the other three elements had no 'text' field
xs, ys, labels, valid_inds, cands, valid_cands = self.vectorize(observations)
if xs is None:
# no valid examples, just return empty responses
return batch_reply
# produce predictions, train on targets if availables
predictions, text_cand_inds, loss = self.predict(xs, ys, cands, valid_cands)
if loss is not None:
if 'metrics' in batch_reply[0]:
for k, v in loss.items():
batch_reply[0]['metrics'][k] = v
else:
batch_reply[0]['metrics'] = loss
if ys is not None:
report_freq = 0
else:
report_freq = 0.1
PaddingUtils.map_predictions(
predictions, valid_inds, batch_reply, observations, self.dict,
self.END_IDX, report_freq=report_freq, labels=labels,
answers=self.answers, ys=ys)
if text_cand_inds is not None:
text_cand_inds = text_cand_inds.cpu().data
for i in range(len(valid_cands)):
order = text_cand_inds[i]
_, batch_idx, curr_cands = valid_cands[i]
curr = batch_reply[batch_idx]
curr['text_candidates'] = [curr_cands[idx] for idx in order
if idx < len(curr_cands)]
return batch_reply
def batch_act(self, observations):
batch_reply = [{'id': self.getID()} for _ in range(len(observations))]
if any(['labels' in obs for obs in observations]):
# if we are starting a new training epoch, reinitialize hidden
if self.is_training == False:
self.hidden = self.model.init_hidden(self.batchsize)
self.is_training = True
data_list, targets_list, _, _, y_lens = self.vectorize(observations, self.opt['seq_len'], self.is_training)
else:
# if we just finished training, reinitialize hidden
if self.is_training == True:
self.hidden = self.model.init_hidden(self.batchsize)
self.is_training = False
data_list, targets_list, labels, valid_inds, y_lens = self.vectorize(observations, self.opt['seq_len'], self.is_training)
if data_list is None:
# not enough data to batch act yet, return empty responses
return batch_reply
batch_reply = []
# during evaluation, len(data_list) is always 1
# during training, len(dat_list) >= 0: vectorize returns a list containing all batches available at the time it is called
for i in range(len(data_list)):
temp_dicts = [{'id': self.getID()} for _ in range(len(observations))]
output, hidden, loss_dict, predictions = self.predict(data_list[i], self.hidden, targets_list[i], self.is_training, y_lens)
self.hidden = self.repackage_hidden(hidden)
if predictions is not None:
# map predictions back to the right order
PaddingUtils.map_predictions(
predictions, valid_inds, temp_dicts, observations,
self.dict, self.END_IDX, report_freq=self.opt['report_freq'])
if loss_dict is not None:
if 'metrics' in temp_dicts[0]:
for k, v in loss_dict.items():
temp_dicts[0]['metrics'][k] = v
else:
temp_dicts[0]['metrics'] = loss_dict
batch_reply += temp_dicts
return batch_reply
def vectorize(self, observations, seq_len, is_training):
"""Convert a list of observations into input & target tensors."""
labels = None
valid_inds = None
y_lens = None
if is_training:
for obs in observations:
if obs:
if 'text2vec' in obs:
self.next_batch += obs['text2vec']
if len(self.next_batch) <= self.batchsize:
return None, None, None, None, None
else:
data_list = []
targets_list = []
# total is the number of batches
total = len(self.next_batch) // self.batchsize
for _ in range(total):
batch = self.next_batch[:self.batchsize]
self.next_batch = self.next_batch[self.batchsize:]
source = torch.LongTensor(batch).t().contiguous()
data = Variable(source[:seq_len])
targets = Variable(source[1:])
if self.use_cuda:
data = data.cuda()
targets = targets.cuda()
data_list.append(data)
targets_list.append(targets)
else:
# here we get valid examples and pad them with zeros
xs, ys, labels, valid_inds, _, y_lens = PaddingUtils.pad_text(
observations,
self.dict,
end_idx=self.END_IDX,
null_idx=self.NULL_IDX)
if self.use_cuda:
if xs is not None:
xs = Variable(torch.LongTensor(xs)).cuda()
if ys is not None:
ys = Variable(torch.LongTensor(ys)).cuda()
else:
if xs is not None:
xs = Variable(torch.LongTensor(xs))
if ys is not None:
ys = Variable(torch.LongTensor(ys))
data_list = [xs]
targets_list = [ys]
return data_list, targets_list, labels, valid_inds, y_lens
def vectorize(self, observations, seq_len, is_training):
"""Convert a list of observations into input & target tensors."""
labels = None
valid_inds = None
y_lens = None
if is_training:
for obs in observations:
if obs:
if 'text2vec' in obs:
self.next_batch += obs['text2vec']
if len(self.next_batch) <= self.batchsize:
return None, None, None, None, None
else:
data_list = []
targets_list = []
# total is the number of batches
total = len(self.next_batch)//self.batchsize
for i in range(total):
batch = self.next_batch[:self.batchsize]
self.next_batch = self.next_batch[self.batchsize:]
source = torch.LongTensor(batch).t().contiguous()
data = Variable(source[:seq_len])
targets = Variable(source[1:])
if self.use_cuda:
data = data.cuda()
targets = targets.cuda()
data_list.append(data)
targets_list.append(targets)
else:
# here we get valid examples and pad them with zeros
xs, ys, labels, valid_inds, _, y_lens = PaddingUtils.pad_text(
observations, self.dict, self.END_IDX, self.NULL_IDX)
if self.use_cuda:
xs = Variable(xs).cuda()
ys = Variable(ys).cuda()
else:
xs = Variable(xs)
ys = Variable(ys)
data_list = [xs]
targets_list = [ys]
return data_list, targets_list, labels, valid_inds, y_lens
def vectorize(self, observations):
"""Convert a list of observations into input & target tensors."""
ys = None
xs, ys, labels, valid_inds, _, _ = PaddingUtils.pad_text(
observations,
self.dict,
self.END_IDX,
self.NULL_IDX,
dq=True,
eval_labels=False,
truncate=self.truncate)
if xs is None:
return None, None, None, None, None, None
if self.use_cuda:
# copy to gpu
self.xs.resize_(xs.size())
self.xs.copy_(xs, async=True)
xs = Variable(self.xs)
if ys is not None:
self.ys.resize_(ys.size())
self.ys.copy_(ys, async=True)
ys = Variable(self.ys)
else:
xs = Variable(xs)
if ys is not None:
ys = Variable(ys)
# set up candidates
cands = None
valid_cands = None
if ys is None and self.rank:
# only do ranking when no targets available and ranking flag set
parsed_cs = []
valid_cands = []
for i, v in enumerate(valid_inds):
if 'label_candidates' in observations[v]:
# each candidate tuple is a pair of the parsed version and
# the original full string
cs = list(observations[v]['label_candidates'])
curr_dqs = [deque(maxlen=self.truncate) for _ in cs]
for dq, c in zip(curr_dqs, cs):
dq.extendleft(reversed(self.parse(c)))
parsed_cs.append(curr_dqs)
valid_cands.append((i, v, cs))
if len(parsed_cs) > 0:
# TODO: store lengths of cands separately, so don't have zero
# padding for varying number of cands per example
# found cands, pack them into tensor
max_c_len = max(max(len(c) for c in cs) for cs in parsed_cs)
max_c_cnt = max(len(cs) for cs in parsed_cs)
for cs in parsed_cs:
for c in cs:
c += [self.NULL_IDX] * (max_c_len - len(c))
cs += [self.NULL_IDX] * (max_c_cnt - len(cs))
cands = torch.LongTensor(parsed_cs)
if self.use_cuda:
# copy to gpu
self.cands.resize_(cands.size())
self.cands.copy_(cands, async=True)
cands = Variable(self.cands)
else:
cands = Variable(cands)
return xs, ys, labels, valid_inds, cands, valid_cands
