def forward(self, words, labels=None):
multitask = labels is not None
if self.training:
words = self.word_vocab.unkify(words)
rnn = self.rnn_builder.initial_state()
word_ids = [self.word_vocab.index_or_unk(word)
for word in [START] + words + [STOP]]
prev_embeddings = [self.embeddings[word_id] for word_id in word_ids[:-1]]
lstm_outputs = rnn.transduce(prev_embeddings)
logits = self.out(dy.concatenate_to_batch(lstm_outputs))
nlls = dy.pickneglogsoftmax_batch(logits, word_ids[1:])
word_nll = dy.sum_batches(nlls)
if multitask:
label_ids = [self.label_vocab.index(label) for label in labels]
logits = self.f_label(dy.concatenate_to_batch(lstm_outputs[1:]))
nlls = dy.pickneglogsoftmax_batch(logits, label_ids)
label_nll = dy.sum_batches(nlls)
# easy proxy to track progress on this task
self.correct += np.sum(np.argmax(logits.npvalue(), axis=0) == label_ids)
self.predicted += len(label_ids)
nll = word_nll + label_nll
else:
nll = word_nll
return nll
def calc_sent_loss(sent):
# Create a computation graph
dy.renew_cg()
# Get embeddings for the sentence
emb = [W_w_p[x] for x in sent]
# Sample K negative words for each predicted word at each position
all_neg_words = np.random.choice(nwords,
size=2 * N * K * len(emb),
replace=True,
p=word_probabilities)
# W_w = dy.parameter(W_w_p)
# Step through the sentence and calculate the negative and positive losses
all_losses = []
for i, my_emb in enumerate(emb):
neg_words = all_neg_words[i * K * 2 * N:(i + 1) * K * 2 * N]
pos_words = (
[sent[x] if x >= 0 else S for x in range(i - N, i)] +
[sent[x] if x < len(sent) else S for x in range(i + 1, i + N + 1)])
neg_loss = -dy.log(
dy.logistic(
-dy.dot_product(my_emb, dy.lookup_batch(W_c_p, neg_words))))
pos_loss = -dy.log(
dy.logistic(
dy.dot_product(my_emb, dy.lookup_batch(W_c_p, pos_words))))
all_losses.append(dy.sum_batches(neg_loss) + dy.sum_batches(pos_loss))
return dy.esum(all_losses)
def compute(self, comb_method: str = "sum") -> Tuple[dy.Expression, Dict]:
"""
Compute loss as DyNet expression by summing over factors and batch elements.
Args:
comb_method: method for combining loss across batch elements ('sum' or 'avg').
Returns:
Scalar DyNet expression.
"""
loss_exprs = 0
loss_data = {}
for name, loss_expr in self.expr_factors.items():
expr, units = loss_expr.loss_value()
loss_exprs += expr
loss_data[name] = dy.sum_batches(expr).value(), units
# Combining
if comb_method == "sum":
loss_exprs = dy.sum_batches(loss_exprs)
elif comb_method == "avg":
loss_exprs = dy.sum_batches(loss_exprs) * (1.0 /
loss_exprs.dim()[1])
else:
raise ValueError(
f"Unknown batch combination method '{comb_method}', expected 'sum' or 'avg'.'"
)
return loss_exprs, loss_data
def _combine_batches(self, batched_expr, comb_method: str = "sum"):
if comb_method == "sum":
return dy.sum_batches(batched_expr)
elif comb_method == "avg":
return dy.sum_batches(batched_expr) * (1.0 / batched_expr.dim()[1])
else:
raise ValueError(f"Unknown batch combination method '{comb_method}', expected 'sum' or 'avg'.'")
def forward(self, words, spans=None):
multitask = spans is not None
if self.training:
words = self.word_vocab.unkify(words)
rnn = self.rnn_builder.initial_state()
word_ids = [self.word_vocab.index_or_unk(word)
for word in [START] + words + [STOP]]
prev_embeddings = [self.embeddings[word_id] for word_id in word_ids[:-1]]
lstm_outputs = rnn.transduce(prev_embeddings)
logits = self.out(dy.concatenate_to_batch(lstm_outputs))
nlls = dy.pickneglogsoftmax_batch(logits, word_ids[1:])
word_nll = dy.sum_batches(nlls)
if multitask:
# predict label for each possible span (null for nonexistent spans)
if self.predict_all_spans:
gold_spans = {(left, right): self.label_vocab.index(label)
for left, right, label in spans}
all_spans = [(left, left + length)
for length in range(1, len(words) + 1)
for left in range(0, len(words) + 1 - length)]
label_ids = [gold_spans.get((left, right), self.label_vocab.size) # last index is for null label
for left, right in all_spans]
# 'lstm minus' features, same as those of the crf parser
span_encodings = [lstm_outputs[right] - lstm_outputs[left]
for left, right in all_spans]
# only predict labels for existing spans
else:
label_ids = [self.label_vocab.index(label) for _, _, label in spans]
# 'lstm minus' features, same as those of the crf parser
span_encodings = [lstm_outputs[right] - lstm_outputs[left]
for left, right, label in spans]
logits = self.f_label(dy.concatenate_to_batch(span_encodings))
nlls = dy.pickneglogsoftmax_batch(logits, label_ids)
label_nll = dy.sum_batches(nlls)
# easy proxy to track progress on this task
self.correct += np.sum(np.argmax(logits.npvalue(), axis=0) == label_ids)
self.predicted += len(label_ids)
nll = word_nll + label_nll
else:
nll = word_nll
return nll
def calc_loss(self, policy):
if self.weight < 1e-8:
return None
neg_entropy = []
for i, ll in enumerate(policy):
if self.valid_pos is not None:
ll = dy.pick_batch_elems(ll, self.valid_pos[i])
loss = dy.sum_batches(dy.sum_elems(dy.cmult(dy.exp(ll), ll)))
neg_entropy.append(dy.sum_batches(loss))
return self.weight * dy.esum(neg_entropy)
def decode_loss(self, src_encodings, tgt_seqs):
"""
:param tgt_seqs: (tgt_heads, tgt_labels): list (length=batch_size) of (src_len)
"""
# todo(NOTE): Sentences should start with empty token (as root of dependency tree)!
tgt_heads, tgt_labels = tgt_seqs
src_len = len(tgt_heads[0])
batch_size = len(tgt_heads)
np_tgt_heads = np.array(tgt_heads).flatten() # (src_len * batch_size)
np_tgt_labels = np.array(tgt_labels).flatten()
s_arc, s_label = self.cal_scores(src_encodings) # (src_len, src_len, bs), ([(src_len, src_len, bs)])
s_arc_value = s_arc.npvalue()
s_arc_choice = np.argmax(s_arc_value, axis=0).transpose().flatten() # (src_len * batch_size)
s_pick_labels = [dy.pick_batch(dy.reshape(score, (src_len,), batch_size=src_len * batch_size), s_arc_choice)
for score in s_label]
s_argmax_labels = dy.concatenate(s_pick_labels, d=0) # n_labels, src_len * batch_size
reshape_s_arc = dy.reshape(s_arc, (src_len,), batch_size=src_len * batch_size)
arc_loss = dy.pickneglogsoftmax_batch(reshape_s_arc, np_tgt_heads)
label_loss = dy.pickneglogsoftmax_batch(s_argmax_labels, np_tgt_labels)
loss = dy.sum_batches(arc_loss + label_loss) / batch_size
return loss
def decode_loss(self, src_encodings, tgt_seqs):
"""
:param tgt_seqs: (tgt_heads, tgt_labels): list (length=batch_size) of (src_len)
"""
# todo(NOTE): Sentences should start with empty token (as root of dependency tree)!
tgt_heads, tgt_labels = tgt_seqs
src_len = len(tgt_heads[0])
batch_size = len(tgt_heads)
np_tgt_heads = np.array(tgt_heads).flatten() # (src_len * batch_size)
np_tgt_labels = np.array(tgt_labels).flatten()
s_arc, s_label = self.cal_scores(src_encodings) # (src_len, src_len, bs), ([(src_len, src_len, bs)])
s_arc_value = s_arc.npvalue()
s_arc_choice = np.argmax(s_arc_value, axis=0).transpose().flatten() # (src_len * batch_size)
s_pick_labels = [dy.pick_batch(dy.reshape(score, (src_len,), batch_size=src_len * batch_size), s_arc_choice)
for score in s_label]
s_argmax_labels = dy.concatenate(s_pick_labels, d=0) # n_labels, src_len * batch_size
reshape_s_arc = dy.reshape(s_arc, (src_len,), batch_size=src_len * batch_size)
arc_loss = dy.pickneglogsoftmax_batch(reshape_s_arc, np_tgt_heads)
label_loss = dy.pickneglogsoftmax_batch(s_argmax_labels, np_tgt_labels)
loss = dy.sum_batches(arc_loss + label_loss) / batch_size
return loss
def fit(self,
X_train=None,
y_train=None,
X_test=None,
y_test=None,
epochs=None):
X_train, y_train, X_test, y_test = self._prepare_data(
X_train, y_train, X_test, y_test)
self._initialize_model()
print('Starting training...')
self.trainer = dy.AdamTrainer(self.model)
losses = []
for _ in tqdm(range(epochs)):
curr_loss = 0.0
for X, y in zip(X_train, y_train):
y_prob = self._predict_proba(X)
loss = dy.sum_batches(dy.pickneglogsoftmax_batch(y_prob, y))
curr_loss += loss.value()
loss.backward()
self.trainer.update()
losses.append(curr_loss / len(X_train))
print('Train Loss:', losses[-1])
self.evaluate(X_test, y_test)
print()
print('Done training')
def compose(
self, embeds: Union[dy.Expression,
List[dy.Expression]]) -> dy.Expression:
if type(embeds) != list:
return dy.sum_batches(embeds)
else:
return dy.esum(embeds)
def calc_loss(self, src, db_idx, src_mask=None, trg_mask=None):
src_embeddings = self.src_embedder.embed_sent(src, mask=src_mask)
self.src_encoder.set_input(src)
src_encodings = self.exprseq_pooling(self.src_encoder.transduce(src_embeddings))
trg_batch, trg_mask = self.database[db_idx]
# print("trg_mask=\n",trg_mask)
trg_encodings = self.encode_trg_example(trg_batch, mask=trg_mask)
dim = trg_encodings.dim()
trg_reshaped = dy.reshape(trg_encodings, (dim[0][0], dim[1]))
# ### DEBUG
# trg_npv = trg_reshaped.npvalue()
# for i in range(dim[1]):
# print("--- trg_reshaped {}: {}".format(i,list(trg_npv[:,i])))
# ### DEBUG
prod = dy.transpose(src_encodings) * trg_reshaped
# ### DEBUG
# prod_npv = prod.npvalue()
# for i in range(dim[1]):
# print("--- prod {}: {}".format(i,list(prod_npv[0].transpose()[i])))
# ### DEBUG
id_range = list(range(len(db_idx)))
# This is ugly:
if self.loss_direction == "forward":
prod = dy.transpose(prod)
loss = dy.sum_batches(dy.hinge_batch(prod, id_range))
elif self.loss_direction == "bidirectional":
prod = dy.reshape(prod, (len(db_idx), len(db_idx)))
loss = dy.sum_elems(
dy.hinge_dim(prod, id_range, d=0) + dy.hinge_dim(prod, id_range, d=1))
else:
raise RuntimeError("Illegal loss direction {}".format(self.loss_direction))
return loss
def test_inputTensor_batched_list(self):
for i in range(4):
dy.renew_cg()
input_tensor = self.input_vals.reshape(self.shapes[i])
xb = dy.inputTensor([np.asarray(x).transpose()
for x in input_tensor.transpose()])
self.assertEqual(
xb.dim()[0],
(self.shapes[i][:-1] if i > 0 else (1,)),
msg="Dimension mismatch"
)
self.assertEqual(
xb.dim()[1],
self.shapes[i][-1],
msg="Dimension mismatch"
)
self.assertTrue(
np.allclose(xb.npvalue(), input_tensor),
msg="Expression value different from initial value"
)
self.assertEqual(
dy.sum_batches(dy.squared_norm(xb)).scalar_value(),
self.squared_norm,
msg="Value mismatch"
)
def calc_loss(self, bisents):
"""
:param bisents: List of (batch size) parallel sentences.
:return: Average batched loss and number of words processed.
"""
dy.renew_cg()
src_sents = [x[0] for x in bisents]
tgt_sents = [x[1] for x in bisents]
self.encode(src_sents)
self.decoder.init(
dy.affine_transform([
dy.parameter(self.b_bridge),
dy.parameter(self.W_bridge),
self.encoder.final_state()
]))
# mask batch
tgt_sents_by_words, masks, num_words = prepare_masks(
tgt_sents, self.tgt_vocab.eos)
prev_words = tgt_sents_by_words[0]
all_losses = []
for next_words, mask in zip(tgt_sents_by_words[1:], masks):
scores = self.decode(
prev_words) # get the decoder output on the previous time step
loss = self.cross_entropy_loss(scores, next_words)
mask_expr = dy.reshape(dy.inputVector(mask), (1, ),
len(bisents)) # change dimension
mask_loss = loss * mask_expr
all_losses.append(mask_loss)
prev_words = next_words
return dy.sum_batches(dy.esum(all_losses)), num_words
def get_loss_batch(self, sent_array):
renew_cg()
init_state = self.builder.initial_state()
R = parameter(self.R)
bias = parameter(self.bias)
wids = []
masks = []
# get the wids and masks for each step
# "I am good", "This is good", "Good Morning" -> [['I', 'Today', 'Good'], ['am', 'is', 'Morning'], ['good', 'good', '<S>'], ['I', 'Today', 'Good'], ['am', 'is', 'Morning'], ['good', 'good', '<S>']]
tot_words = 0
wids = []
masks = []
for i in range(len(sent_array[0])):
wids.append([(sent[i] if len(sent) > i else 3)
for sent in sent_array])
mask = [(1 if len(sent) > i else 0) for sent in sent_array]
masks.append(mask)
tot_words += sum(mask)
# start the rnn by inputting "<s>"
init_ids = [2] * len(sent_array)
#print dy.lookup_batch(self.lookup,init_ids)
#print "Looked up"
s = init_state.add_input(dy.lookup_batch(self.lookup, init_ids))
# feed word vectors into the RNN and predict the next word
losses = []
for wid, mask in zip(wids, masks):
# calculate the softmax and loss
#print "WID ", wid
score = dy.affine_transform([bias, R, s.output()])
loss = dy.pickneglogsoftmax_batch(score, wid)
# mask the loss if at least one sentence is shorter
if mask[-1] != 1:
mask_expr = dy.inputVector(mask)
mask_expr = dy.reshape(mask_expr, (1, ), len(sent_array))
loss = loss * mask_expr
losses.append(loss)
# update the state of the RNN
wemb = dy.lookup_batch(self.lookup, wid)
s = s.add_input(wemb)
return dy.sum_batches(dy.esum(losses)), tot_words
errs = [] # will hold expressions
es = []
for (wid, mask) in zip(wids, masks):
# assume word is already a word-id
x_t = lookup(self.lookup, int(cw))
state = state.add_input(x_t)
y_t = state.output()
r_t = bias + (R * y_t)
err = pickneglogsoftmax(r_t, int(nw))
errs.append(err)
nerr = esum(errs)
return nerr
def calc_loss(self, policy_reward, only_final_reward=True):
loss = losses.FactoredLossExpr()
## Calculate baseline
pred_reward, baseline_loss = self.calc_baseline_loss(policy_reward, only_final_reward)
if only_final_reward:
rewards = [policy_reward - pw_i for pw_i in pred_reward]
else:
rewards = [pr_i - pw_i for pr_i, pw_i in zip(policy_reward, pred_reward)]
loss.add_loss("rl_baseline", baseline_loss)
## Z-Normalization
rewards = dy.concatenate(rewards, d=0)
if self.z_normalization:
rewards_value = rewards.value()
rewards_mean = np.mean(rewards_value)
rewards_std = np.std(rewards_value) + 1e-10
rewards = (rewards - rewards_mean) / rewards_std
## Calculate Confidence Penalty
if self.confidence_penalty:
cp_loss = self.confidence_penalty.calc_loss(self.policy_lls)
loss.add_loss("rl_confpen", cp_loss)
## Calculate Reinforce Loss
reinf_loss = []
# Loop through all action in one sequence
for i, (policy, action) in enumerate(zip(self.policy_lls, self.actions)):
# Main Reinforce calculation
reward = dy.pick(rewards, i)
ll = dy.pick_batch(policy, action)
if self.valid_pos is not None:
ll = dy.pick_batch_elems(ll, self.valid_pos[i])
reward = dy.pick_batch_elems(reward, self.valid_pos[i])
reinf_loss.append(dy.sum_batches(ll * reward))
loss.add_loss("rl_reinf", -self.weight * dy.esum(reinf_loss))
## the composed losses
return loss
def get_loss(x, y):
"""
Get loss -log(softmax(score[y]))
"""
score = run_MLP(x)
bsize = x.shape[0]
return dy.sum_batches(dy.pickneglogsoftmax_batch(score, y)) / bsize
def add_loss(self, loss_name, loss_expr):
if type(loss_expr) == LossBuilder:
self.loss_nodes.extend(loss_expr.loss_nodes)
else:
if loss_expr.dim()[1] > 1:
loss_expr = dy.sum_batches(loss_expr)
self.loss_nodes.append((loss_name, loss_expr))
def get_loss(x, y):
"""
Get loss -log(softmax(score[y]))
"""
score = run_IRNN(x)
bsize, seq_len = x.shape
return dy.sum_batches(dy.pickneglogsoftmax_batch(score, y)) / bsize
def calc_loss(self, src_seqs, trg_seqs, training=True):
batch_size = len(src_seqs)
src_encodings = self.encoder.encode(src_seqs, training=training)
src_enc_all = dy.concatenate_cols(src_encodings)
src_trans_att = self.attender.get_src_transformation(src_enc_all)
state = self.decoder.initialize(src_encodings, training=training)
ctx_tm1 = dy.vecInput(self.encoder.state_dim)
losses = []
max_len = max(map(len, trg_seqs))
for i in xrange(1, max_len):
y_tm1 = [trg_seq[i - 1] if i < len(trg_seq) else trg_seq[-1] for trg_seq in trg_seqs]
ref_y_t = [trg_seq[i] if i < len(trg_seq) else trg_seq[-1] for trg_seq in trg_seqs]
y_tm1_embed = self.decoder.embedder.embed_item(y_tm1, training=training)
x = dy.concatenate([y_tm1_embed, ctx_tm1])
state = state.add_input(x)
h_t = state.output()
ctx_t, alpha_t = self.attender.calc_context(src_enc_all, src_trans_att, h_t)
loss_t = self.decoder.calc_loss(h_t, ctx_t, ref_y_t, training=training)
mask = dy.inputVector([1 if i < len(trg_seq) else 0 for trg_seq in trg_seqs])
mask = dy.reshape(mask, (1,), batch_size)
loss_t = dy.sum_batches(loss_t * mask)
ctx_tm1 = ctx_t
losses.append(loss_t)
loss = dy.esum(losses)
return loss
def step_batch(self, instances):
dy.renew_cg()
self.l2r_builder.set_dropout(0.2)
self.r2l_builder.set_dropout(0.2)
self.dec_builder.set_dropout(0.2)
W_y = dy.parameter(self.W_y)
b_y = dy.parameter(self.b_y)
src_sents = [x[0] for x in instances]
padded_src = self.__pad_batch(src_sents, True)
src_cws = np.transpose(padded_src)
tgt_sents = [x[1] for x in instances]
padded_tgt = self.__pad_batch(tgt_sents, False)
masks_tgt, num_words = self.__mask(tgt_sents)
masks_tgt = np.transpose(masks_tgt)
padded_tgt = np.transpose(padded_tgt)
instance_size = len(instances)
src_cws_rev = list(reversed(src_cws))
# Bidirectional representations
l2r_state = self.l2r_builder.initial_state()
r2l_state = self.r2l_builder.initial_state()
l2r_contexts = []
r2l_contexts = []
for (cws_l2r, cws_r2l) in zip(src_cws, src_cws_rev):
l2r_state = l2r_state.add_input(dy.lookup_batch(self.src_lookup, cws_l2r))
r2l_state = r2l_state.add_input(dy.lookup_batch(self.src_lookup, cws_r2l))
l2r_contexts.append(l2r_state.output()) #[<S>, x_1, x_2, ..., </S>]
r2l_contexts.append(r2l_state.output()) #[</S> x_n, x_{n-1}, ... <S>]
r2l_contexts.reverse() #[<S>, x_1, x_2, ..., </S>]
# Combine the left and right representations for every word
h_fs = []
for (l2r_i, r2l_i) in zip(l2r_contexts, r2l_contexts):
h_fs.append(dy.concatenate([l2r_i, r2l_i]))
h_fs_matrix = dy.concatenate_cols(h_fs)
losses = []
# Decoder
c_t = dy.vecInput(self.hidden_size * 2)
start = dy.concatenate([dy.lookup_batch(self.tgt_lookup, len(tgt_sents) * [self.tgt_token_to_id[self.src_pad]]), c_t])
dec_state = self.dec_builder.initial_state().add_input(start)
for (cws, nws, mask) in zip(padded_tgt, padded_tgt[1:], masks_tgt):
h_e = dec_state.output()
c_t = self.__attention_mlp(h_fs_matrix, h_e)
# Get the embedding for the current target word
embed_t = dy.lookup_batch(self.tgt_lookup, cws)
# Create input vector to the decoder
x_t = dy.concatenate([embed_t, c_t])
dec_state = dec_state.add_input(x_t)
y_star = b_y + W_y * dec_state.output()
loss = dy.pickneglogsoftmax_batch(y_star, nws)
if mask[-1] == 0:
mask_loss = dy.reshape(dy.inputVector(mask), (1,), instance_size)
masked = loss * mask_loss
losses.append(masked)
else:
losses.append(loss)
#losses = [(x / num_words) for x in losses]
return dy.sum_batches(dy.esum(losses)), num_words
def compute_decoder_batch_loss(self, encoded_inputs, input_masks, output_word_ids, output_masks, batch_size):
self.readout = dn.parameter(self.params['readout'])
self.bias = dn.parameter(self.params['bias'])
self.w_c = dn.parameter(self.params['w_c'])
self.u_a = dn.parameter(self.params['u_a'])
self.v_a = dn.parameter(self.params['v_a'])
self.w_a = dn.parameter(self.params['w_a'])
# initialize the decoder rnn
s_0 = self.decoder_rnn.initial_state()
# initial "input feeding" vectors to feed decoder - 3*h
init_input_feeding = dn.lookup_batch(self.init_lookup, [0] * batch_size)
# initial feedback embeddings for the decoder, use begin seq symbol embedding
init_feedback = dn.lookup_batch(self.output_lookup, [self.y2int[common.BEGIN_SEQ]] * batch_size)
# init decoder rnn
decoder_init = dn.concatenate([init_feedback, init_input_feeding])
s = s_0.add_input(decoder_init)
# loss per timestep
losses = []
# run the decoder through the output sequences and aggregate loss
for i, step_word_ids in enumerate(output_word_ids):
# returns h x batch size matrix
decoder_rnn_output = s.output()
# compute attention context vector for each sequence in the batch (returns 2h x batch size matrix)
attention_output_vector, alphas = self.attend(encoded_inputs, decoder_rnn_output, input_masks)
# compute output scores (returns vocab_size x batch size matrix)
# h = readout * attention_output_vector + bias
h = dn.affine_transform([self.bias, self.readout, attention_output_vector])
# get batch loss for this timestep
batch_loss = dn.pickneglogsoftmax_batch(h, step_word_ids)
# mask the loss if at least one sentence is shorter
if output_masks and output_masks[i][-1] != 1:
mask_expr = dn.inputVector(output_masks[i])
# noinspection PyArgumentList
mask_expr = dn.reshape(mask_expr, (1,), batch_size)
batch_loss = batch_loss * mask_expr
# input feeding approach - input h (attention_output_vector) to the decoder
# prepare for the next iteration - "feedback"
feedback_embeddings = dn.lookup_batch(self.output_lookup, step_word_ids)
decoder_input = dn.concatenate([feedback_embeddings, attention_output_vector])
s = s.add_input(decoder_input)
losses.append(batch_loss)
# sum the loss over the time steps and batch
total_batch_loss = dn.sum_batches(dn.esum(losses))
return total_batch_loss
def step_batch(self, instances):
dy.renew_cg()
W_y = dy.parameter(self.W_y)
b_y = dy.parameter(self.b_y)
src_sents = [x[0] for x in instances]
padded_src = self.__pad_batch(src_sents)
masks_src = np.transpose(self.__mask(padded_src))
src_cws = np.transpose(padded_src)
tgt_sents = [x[1] for x in instances]
tgt_ids = []
for sent in tgt_sents:
sent = [self.tgt_token_to_id[x] for x in sent]
tgt_ids.append(sent)
tgt_ids = map(list, zip(*tgt_ids))
padded_src_rev = list(reversed(padded_src))
src_cws_rev = np.transpose(padded_src_rev)
# Bidirectional representations
l2r_state = self.l2r_builder.initial_state()
r2l_state = self.r2l_builder.initial_state()
l2r_contexts = []
r2l_contexts = []
for (cws_l2r, cws_r2l) in zip(src_cws, src_cws_rev):
l2r_state = l2r_state.add_input(dy.lookup_batch(self.src_lookup, cws_l2r))
r2l_state = r2l_state.add_input(dy.lookup_batch(self.src_lookup, cws_r2l))
l2r_contexts.append(l2r_state.output()) #[<S>, x_1, x_2, ..., </S>]
r2l_contexts.append(r2l_state.output()) #[</S> x_n, x_{n-1}, ... <S>]
r2l_contexts.reverse() #[<S>, x_1, x_2, ..., </S>]
# Combine the left and right representations for every word
h_fs = []
for (l2r_i, r2l_i) in zip(l2r_contexts, r2l_contexts):
h_fs.append(dy.concatenate([l2r_i, r2l_i]))
h_fs_matrix = dy.concatenate_cols(h_fs)
losses = []
num_words = 0
# Decoder
c_t = dy.vecInput(self.hidden_size * 2)
start = dy.concatenate([dy.lookup_batch(self.tgt_lookup, len(tgt_sents) * [self.tgt_token_to_id['<S>']]), c_t])
dec_state = self.dec_builder.initial_state().add_input(start)
for (cws, nws, mask) in zip(tgt_ids, tgt_ids[1:], masks_src):
h_e = dec_state.output()
c_t = self.__attention_mlp(h_fs_matrix, h_e)
# Get the embedding for the current target word
embed_t = dy.lookup_batch(self.tgt_lookup, cws)
# Create input vector to the decoder
x_t = dy.concatenate([embed_t, c_t])
dec_state = dec_state.add_input(x_t)
y_star = b_y + W_y * dec_state.output()
loss = dy.pickneglogsoftmax_batch(y_star, nws)
if mask[0] == 0:
mask_loss = dy.reshape(dy.inputVector(mask), (1,), self.BATCH_SIZE)
loss = loss * mask_loss
losses.append(loss)
num_words += 1
return dy.sum_batches(dy.esum(losses)/num_words), num_words
def calc_loss(sents):
dy.renew_cg()
# Transduce all batch elements with an LSTM
src_sents = [x[0] for x in sents]
tgt_sents = [x[1] for x in sents]
src_cws = []
src_len = [len(sent) for sent in src_sents]
max_src_len = np.max(src_len)
num_words = 0
for i in range(max_src_len):
src_cws.append([sent[i] for sent in src_sents])
#initialize the LSTM
init_state_src = LSTM_SRC_BUILDER.initial_state()
#get the output of the first LSTM
src_output = init_state_src.add_inputs(
[dy.lookup_batch(LOOKUP_SRC, cws) for cws in src_cws])[-1].output()
#now decode
all_losses = []
# Decoder
#need to mask padding at end of sentence
tgt_cws = []
tgt_len = [len(sent) for sent in sents]
max_tgt_len = np.max(tgt_len)
masks = []
for i in range(max_tgt_len):
tgt_cws.append(
[sent[i] if len(sent) > i else eos_trg for sent in tgt_sents])
mask = [(1 if len(sent) > i else 0) for sent in tgt_sents]
masks.append(mask)
num_words += sum(mask)
current_state = LSTM_TRG_BUILDER.initial_state().set_s(
[src_output, dy.tanh(src_output)])
prev_words = tgt_cws[0]
W_sm = dy.parameter(W_sm_p)
b_sm = dy.parameter(b_sm_p)
for next_words, mask in zip(tgt_cws[1:], masks):
#feed the current state into the
current_state = current_state.add_input(
dy.lookup_batch(LOOKUP_TRG, prev_words))
output_embedding = current_state.output()
s = dy.affine_transform([b_sm, W_sm, output_embedding])
loss = (dy.pickneglogsoftmax_batch(s, next_words))
mask_expr = dy.inputVector(mask)
mask_expr = dy.reshape(mask_expr, (1, ), len(sents))
mask_loss = loss * mask_expr
all_losses.append(mask_loss)
prev_words = next_words
return dy.sum_batches(dy.esum(all_losses)), num_words
def calc_loss(sents):
dy.renew_cg()
# Transduce all batch elements with an LSTM
src_sents = [x[0] for x in sents]
tgt_sents = [x[1] for x in sents]
src_cws = []
src_len = [len(sent) for sent in src_sents]
max_src_len = np.max(src_len)
num_words = 0
for i in range(max_src_len):
src_cws.append([sent[i] for sent in src_sents])
#initialize the LSTM
init_state_src = LSTM_SRC_BUILDER.initial_state()
#get the output of the first LSTM
src_output = init_state_src.add_inputs([dy.lookup_batch(LOOKUP_SRC, cws) for cws in src_cws])[-1].output()
#now decode
all_losses = []
# Decoder
#need to mask padding at end of sentence
tgt_cws = []
tgt_len = [len(sent) for sent in sents]
max_tgt_len = np.max(tgt_len)
masks = []
for i in range(max_tgt_len):
tgt_cws.append([sent[i] if len(sent) > i else eos_trg for sent in tgt_sents])
mask = [(1 if len(sent) > i else 0) for sent in tgt_sents]
masks.append(mask)
num_words += sum(mask)
current_state = LSTM_TRG_BUILDER.initial_state().set_s([src_output, dy.tanh(src_output)])
prev_words = tgt_cws[0]
W_sm = dy.parameter(W_sm_p)
b_sm = dy.parameter(b_sm_p)
for next_words, mask in zip(tgt_cws[1:], masks):
#feed the current state into the
current_state = current_state.add_input(dy.lookup_batch(LOOKUP_TRG, prev_words))
output_embedding = current_state.output()
s = dy.affine_transform([b_sm, W_sm, output_embedding])
loss = (dy.pickneglogsoftmax_batch(s, next_words))
mask_expr = dy.inputVector(mask)
mask_expr = dy.reshape(mask_expr, (1,),len(sents))
mask_loss = loss * mask_expr
all_losses.append(mask_loss)
prev_words = next_words
return dy.sum_batches(dy.esum(all_losses)), num_words
def compute_loss(self, words, extwords, tags, true_arcs, true_labels):
arc_logits, rel_logits = self.forward(words, extwords, tags, True)
seq_len = len(true_arcs)
targets_1D = dynet_flatten_numpy(true_arcs)
flat_arc_logits = dy.reshape(arc_logits, (seq_len, ), seq_len)
losses = dy.pickneglogsoftmax_batch(flat_arc_logits, targets_1D)
arc_loss = dy.sum_batches(losses)
flat_rel_logits = dy.reshape(rel_logits, (seq_len, self.rel_size),
seq_len)
partial_rel_logits = dy.pick_batch(flat_rel_logits, targets_1D)
targets_rel1D = dynet_flatten_numpy(true_labels)
losses = dy.pickneglogsoftmax_batch(partial_rel_logits, targets_rel1D)
rel_loss = dy.sum_batches(losses)
loss = arc_loss + rel_loss
return loss
def calc_loss(self, src, db_idx):
src_embeddings = self.src_embedder.embed_sent(src)
src_encodings = self.exprseq_pooling(
self.src_encoder.transduce(src_embeddings))
trg_encodings = self.encode_trg_example(self.database[db_idx])
prod = dy.transpose(dy.transpose(src_encodings) * trg_encodings)
loss = dy.sum_batches(
dy.hinge_batch(prod, list(six.moves.range(len(db_idx)))))
print(loss.npvalue())
return loss
def calc_sent_loss(sent):
# Create a computation graph
dy.renew_cg()
# Get embeddings for the sentence
emb = [W_w_p[x] for x in sent]
# Sample K negative words for each predicted word at each position
all_neg_words = np.random.choice(nwords, size=2*N*K*len(emb), replace=True, p=word_probabilities)
# W_w = dy.parameter(W_w_p)
# Step through the sentence and calculate the negative and positive losses
all_losses = []
for i, my_emb in enumerate(emb):
neg_words = all_neg_words[i*K*2*N:(i+1)*K*2*N]
pos_words = ([sent[x] if x >= 0 else S for x in range(i-N,i)] +
[sent[x] if x < len(sent) else S for x in range(i+1,i+N+1)])
neg_loss = -dy.log(dy.logistic(-dy.dot_product(my_emb, dy.lookup_batch(W_c_p, neg_words))))
pos_loss = -dy.log(dy.logistic(dy.dot_product(my_emb, dy.lookup_batch(W_c_p, pos_words))))
all_losses.append(dy.sum_batches(neg_loss) + dy.sum_batches(pos_loss))
return dy.esum(all_losses)
def forward(self, inputs, expected_output, droput1=0.0, droput2=0.0):
out = self(inputs, droput1, droput2)
expected_output = np.array([self._T2I[exp] for exp in expected_output])
loss = dy.sum_batches(dy.pickneglogsoftmax_batch(out, expected_output))
predictions_probs = out.npvalue()
if len(predictions_probs.shape) == 2:
predictions = np.argmax(predictions_probs.T, axis=1)
else:
predictions = np.array([np.argmax(predictions_probs)])
predictions = np.array(
[self._I2T[prediction] for prediction in predictions])
return loss, predictions
def test_inputTensor_batched_list(self):
for i in range(4):
dy.renew_cg()
input_tensor = self.input_vals.reshape(self.shapes[i])
xb = dy.inputTensor([np.asarray(x).transpose() for x in input_tensor.transpose()])
self.assertEqual(xb.dim()[0], (self.shapes[i][:-1] if i > 0 else (1,)),
msg="Dimension mismatch")
self.assertEqual(xb.dim()[1], self.shapes[i][-1],
msg="Dimension mismatch")
self.assertTrue(np.allclose(xb.npvalue(), input_tensor),
msg="Expression value different from initial value")
self.assertEqual(dy.sum_batches(dy.squared_norm(xb)).scalar_value(),
self.squared_norm, msg="Value mismatch")
def fit_partial(self, instances):
random.shuffle(instances)
self.iter += 1
losses = []
dy.renew_cg()
total_loss, total_size = 0., 0
prog = tqdm(desc="Epoch {}".format(self.iter), ncols=80, total=len(instances) + 1)
for i, ins in enumerate(instances, 1):
losses.extend(list(self.model.loss(*ins)))
if i % self.batch_size == 0:
loss = dy.sum_batches(dy.concatenate_to_batch(losses))
total_loss += loss.value()
total_size += len(losses)
prog.set_postfix(loss=loss.value()/len(losses))
loss.backward()
self.opt.update()
dy.renew_cg()
losses = []
prog.update()
if losses:
loss = dy.sum_batches(dy.concatenate_to_batch(losses))
total_loss += loss.value()
total_size += len(losses)
self.loss = total_loss / total_size
prog.set_postfix(loss=self.loss)
loss.backward()
self.opt.update()
dy.renew_cg()
prog.update()
self.opt.learning_rate *= self.lr_decay
prog.close()
def sum(x, dim=None, include_batch_dim=False):
if isinstance(x, list):
return dy.esum(x)
head_shape, batch_size = x.dim()
if dim is None:
x = dy.sum_elems(x)
if include_batch_dim and batch_size > 1:
return dy.sum_batches(x)
else:
return x
else:
if dim == -1:
dim = len(head_shape) - 1
return dy.sum_dim(x, d=[dim], b=include_batch_dim)
def calc_sent_loss(sent, dropout=0.0):
# Create a computation graph
dy.renew_cg()
# The initial history is equal to end of sentence symbols
hist = [S] * N
# Step through the sentence, including the end of sentence token
all_histories = []
all_targets = []
for next_word in sent + [S]:
all_histories.append(list(hist))
all_targets.append(next_word)
hist = hist[1:] + [next_word]
s = calc_score_of_histories(all_histories, dropout=dropout)
return dy.sum_batches(dy.pickneglogsoftmax_batch(s, all_targets))
def calc_sent_loss(sent, dropout=0.0):
# Create a computation graph
dy.renew_cg()
# The initial history is equal to end of sentence symbols
hist = [S] * N
# Step through the sentence, including the end of sentence token
all_histories = []
all_targets = []
for next_word in sent + [S]:
all_histories.append(list(hist))
all_targets.append(next_word)
hist = hist[1:] + [next_word]
s = calc_score_of_histories(all_histories, dropout=dropout)
return dy.sum_batches(dy.pickneglogsoftmax_batch(s, all_targets))
def calc_nll(self, src: Union[batchers.Batch, sent.Sentence],
trg: Union[batchers.Batch, sent.Sentence]) -> dy.Expression:
event_trigger.start_sent(src)
if isinstance(src, batchers.CompoundBatch): src = src.batches[0]
# Encode the sentence
initial_state = self._encode_src(src)
dec_state = initial_state
trg_mask = trg.mask if batchers.is_batched(trg) else None
cur_losses = []
seq_len = trg.sent_len()
if settings.CHECK_VALIDITY and batchers.is_batched(src):
for j, single_trg in enumerate(trg):
assert single_trg.sent_len(
) == seq_len # assert consistent length
assert 1 == len([
i for i in range(seq_len)
if (trg_mask is None or trg_mask.np_arr[j, i] == 0)
and single_trg[i] == vocabs.Vocab.ES
]) # assert exactly one unmasked ES token
input_word = None
for i in range(seq_len):
ref_word = DefaultTranslator._select_ref_words(
trg, i, truncate_masked=self.truncate_dec_batches)
if self.truncate_dec_batches and batchers.is_batched(ref_word):
dec_state.rnn_state, ref_word = batchers.truncate_batches(
dec_state.rnn_state, ref_word)
if input_word is not None:
dec_state = self.decoder.add_input(
dec_state, self.trg_embedder.embed(input_word))
rnn_output = dec_state.rnn_state.output()
dec_state.context = self.attender.calc_context(rnn_output)
word_loss = self.decoder.calc_loss(dec_state, ref_word)
if not self.truncate_dec_batches and batchers.is_batched(
src) and trg_mask is not None:
word_loss = trg_mask.cmult_by_timestep_expr(word_loss,
i,
inverse=True)
cur_losses.append(word_loss)
input_word = ref_word
if self.truncate_dec_batches:
loss_expr = dy.esum([dy.sum_batches(wl) for wl in cur_losses])
else:
loss_expr = dy.esum(cur_losses)
return loss_expr
def calc_baseline_loss(self, reward, only_final_reward):
pred_rewards = []
cur_losses = []
for i, state in enumerate(self.states):
pred_reward = self.baseline.transform(dy.nobackprop(state))
pred_rewards.append(dy.nobackprop(pred_reward))
seq_reward = reward if only_final_reward else reward[i]
if self.valid_pos is not None:
pred_reward = dy.pick_batch_elems(pred_reward, self.valid_pos[i])
act_reward = dy.pick_batch_elems(seq_reward, self.valid_pos[i])
else:
act_reward = seq_reward
cur_losses.append(dy.sum_batches(dy.squared_distance(pred_reward, dy.nobackprop(act_reward))))
return pred_rewards, dy.esum(cur_losses)
def learn(self, batch_size):
if self.prioritized:
if not self.memory.is_full(): return -np.inf
indices, exps, weights = self.memory.sample(batch_size, self.beta)
else:
exps = self.memory.sample(batch_size)
obss, actions, rewards, obs_nexts, dones = self._process(exps)
dy.renew_cg()
target_network = self.target_network if self.use_double_dqn else self.network
if self.dueling:
target_values, v = target_network(obs_nexts, batched=True)
target_values = target_values.npvalue() + v.npvalue()
else:
target_values = target_network(obs_nexts, batched=True)
target_values = target_values.npvalue()
target_values = np.max(target_values, axis=0)
target_values = rewards + self.reward_decay * (target_values * (1 - dones))
dy.renew_cg()
if self.dueling:
all_values_expr, v = self.network(obss, batched=True)
else:
all_values_expr = self.network(obss, batched=True)
picked_values = dy.pick_batch(all_values_expr, actions)
diff = (picked_values + v if self.dueling else picked_values) - dy.inputTensor(target_values, batched=True)
if self.prioritized:
self.memory.update(indices, np.transpose(np.abs(diff.npvalue())))
losses = dy.pow(diff, dy.constant(1, 2))
if self.prioritized:
losses = dy.cmult(losses, dy.inputTensor(weights, batched=True))
loss = dy.sum_batches(losses)
loss_value = loss.npvalue()
loss.backward()
self.trainer.update()
self.epsilon = max(self.epsilon - self.epsilon_decrease, self.epsilon_lower)
if self.prioritized:
self.beta = min(self.beta + self.beta_increase, 1.)
self.learn_step += 1
if self.use_double_dqn and self.learn_step % self.n_replace_target == 0:
self.target_network.update(self.network)
return loss_value
def BuildLMGraph(self, sents):
dy.renew_cg()
# initialize the RNN
init_state = self.builder.initial_state()
# parameters -> expressions
R = dy.parameter(self.R)
bias = dy.parameter(self.bias)
S = vocab.w2i["<s>"]
# get the cids and masks for each step
tot_chars = 0
cids = []
masks = []
for i in range(len(sents[0])):
cids.append([(vocab.w2i[sent[i]] if len(sent) > i else S) for sent in sents])
mask = [(1 if len(sent)>i else 0) for sent in sents]
masks.append(mask)
tot_chars += sum(mask)
# start the rnn with "<s>"
init_ids = cids[0]
s = init_state.add_input(lookup_batch(self.lookup, init_ids))
losses = []
# feed char vectors into the RNN and predict the next char
for cid, mask in zip(cids[1:], masks[1:]):
score = dy.affine_transform([bias, R, s.output()])
loss = dy.pickneglogsoftmax_batch(score, cid)
# mask the loss if at least one sentence is shorter
if mask[-1] != 1:
mask_expr = dy.inputVector(mask)
mask_expr = dy.reshape(mask_expr, (1,), len(sents))
loss = loss * mask_expr
losses.append(loss)
# update the state of the RNN
cemb = dy.lookup_batch(self.lookup, cid)
s = s.add_input(cemb)
return dy.sum_batches(dy.esum(losses)), tot_chars
def calc_lm_loss(sents):
dy.renew_cg()
# initialize the RNN
f_init = RNN.initial_state()
# get the wids and masks for each step
tot_words = 0
wids = []
masks = []
for i in range(len(sents[0])):
wids.append([(sent[i] if len(sent) > i else S) for sent in sents])
mask = [(1 if len(sent) > i else 0) for sent in sents]
masks.append(mask)
tot_words += sum(mask)
# start the rnn by inputting "<s>"
init_ids = [S] * len(sents)
s = f_init.add_input(dy.lookup_batch(WORDS_LOOKUP, init_ids))
# feed word vectors into the RNN and predict the next word
losses = []
for wid, mask in zip(wids, masks):
# calculate the softmax and loss
score = dy.affine_transform([b_exp, W_exp, s.output()])
loss = dy.pickneglogsoftmax_batch(score, wid)
# mask the loss if at least one sentence is shorter
if mask[-1] != 1:
mask_expr = dy.inputVector(mask)
mask_expr = dy.reshape(mask_expr, (1,), len(sents))
loss = loss * mask_expr
losses.append(loss)
# update the state of the RNN
wemb = dy.lookup_batch(WORDS_LOOKUP, wid)
s = s.add_input(wemb)
return dy.sum_batches(dy.esum(losses)), tot_words
def calc_loss(sents):
dy.renew_cg()
# Transduce all batch elements with an LSTM
src_sents = [x[0] for x in sents]
tgt_sents = [x[1] for x in sents]
src_cws = []
src_len = [len(sent) for sent in src_sents]
max_src_len = np.max(src_len)
num_words = 0
for i in range(max_src_len):
src_cws.append([sent[i] for sent in src_sents])
#get the outputs of the first LSTM
src_outputs = [dy.concatenate([x.output(), y.output()]) for x,y in LSTM_SRC.add_inputs([dy.lookup_batch(LOOKUP_SRC, cws) for cws in src_cws])]
src_output = src_outputs[-1]
#gets the parameters for the attention
src_output_matrix = dy.concatenate_cols(src_outputs)
w1_att_src = dy.parameter(w1_att_src_p)
fixed_attentional_component = w1_att_src * src_output_matrix
#now decode
all_losses = []
# Decoder
#need to mask padding at end of sentence
tgt_cws = []
tgt_len = [len(sent) for sent in sents]
max_tgt_len = np.max(tgt_len)
masks = []
for i in range(max_tgt_len):
tgt_cws.append([sent[i] if len(sent) > i else eos_trg for sent in tgt_sents])
mask = [(1 if len(sent) > i else 0) for sent in tgt_sents]
masks.append(mask)
num_words += sum(mask)
current_state = LSTM_TRG_BUILDER.initial_state().set_s([src_output, dy.tanh(src_output)])
prev_words = tgt_cws[0]
W_sm = dy.parameter(W_sm_p)
b_sm = dy.parameter(b_sm_p)
W_m = dy.parameter(W_m_p)
b_m = dy.parameter(b_m_p)
for next_words, mask in zip(tgt_cws[1:], masks):
#feed the current state into the
current_state = current_state.add_input(dy.lookup_batch(LOOKUP_TRG, prev_words))
output_embedding = current_state.output()
att_output, _ = calc_attention(src_output_matrix, output_embedding, fixed_attentional_component)
middle_expr = dy.tanh(dy.affine_transform([b_m, W_m, dy.concatenate([output_embedding, att_output])]))
s = dy.affine_transform([b_sm, W_sm, middle_expr])
loss = (dy.pickneglogsoftmax_batch(s, next_words))
mask_expr = dy.inputVector(mask)
mask_expr = dy.reshape(mask_expr, (1,),len(sents))
mask_loss = loss * mask_expr
all_losses.append(mask_loss)
prev_words = next_words
return dy.sum_batches(dy.esum(all_losses)), num_words
