Module pysimt.translators.waitk_greedy
Expand source code
import logging
import time
from .sim_greedy import SimultaneousGreedySearch
from ..utils.data import sort_predictions
from ..utils.device import DEVICE
from ..utils.io import progress_bar
logger = logging.getLogger('pysimt')
class SimultaneousWaitKGreedySearch(SimultaneousGreedySearch):
def run_all(self):
"""Do a grid search over the given list of parameters."""
# Let's pretend that `s_0` is `k`
for k in self.list_of_s_0:
# Run the decoding
hyps, actions, up_time = self.run(int(k))
# Dumps two files one with segmentations preserved, another
# with post-processing filters applied
self.dump_results(hyps, suffix=f'wait{k}')
# Dump actions
self.dump_lines(actions, suffix=f'wait{k}.acts')
def run(self, k):
# R/W actions generated for the whole test set
actions = []
# Final translations
translations = []
start = time.time()
for batch in progress_bar(self.data_loader, unit='batch'):
self.clear_states()
batch.device(DEVICE)
# Compute all at once
self.cache_encoder_states(batch)
# Read some words and get trimmed states
state_dict = self.read_more(k)
self.update_s(k)
# Initial state is None i.e. 0. state_dict is not used
self.prev_h = self.decoder_init(state_dict)
# last batch could be smaller than the requested batch size
cur_batch_size = batch.size
# Start all sentences with <s>
self.set_first_word_to_bos(cur_batch_size)
# We will start by writing
next_action = self.ACT_WRITE
while not self.eos_written and self.t_ptr < self.max_len:
if next_action == self.ACT_WRITE or self.is_src_read():
logp, new_h, new_word = self.decoder_step(
state_dict, self.prev_word, self.prev_h, self.tf_decoder_input)
self.write(new_word, new_h)
else:
# READ
state_dict = self.read_more(1)
self.update_s(1)
# Invert the last committed action for interleaved decoding
next_action = 1 - self.actions[-1]
# All finished, convert translations to python lists on CPU
idxs = self.buffer[self.buffer.ne(0)].tolist()
if idxs[-1] != self.eos:
# In cases where <eos> not produced and the above loop
# went on until max_len, add an explicit <eos> for correctness
idxs.append(self.eos)
# compute action sequence from which metrics will be computed
actions.append(' '.join(map(lambda i: str(i), self.actions)))
translations.append(self.vocab.idxs_to_sent(idxs))
up_time = time.time() - start
hyps = sort_predictions(self.data_loader, translations)
actions = sort_predictions(self.data_loader, actions)
return (hyps, actions, up_time)Classes
class SimultaneousWaitKGreedySearch (model, data_loader, out_prefix, batch_size, filter_chain=None, max_len=100, **kwargs)-
Expand source code
class SimultaneousWaitKGreedySearch(SimultaneousGreedySearch): def run_all(self): """Do a grid search over the given list of parameters.""" # Let's pretend that `s_0` is `k` for k in self.list_of_s_0: # Run the decoding hyps, actions, up_time = self.run(int(k)) # Dumps two files one with segmentations preserved, another # with post-processing filters applied self.dump_results(hyps, suffix=f'wait{k}') # Dump actions self.dump_lines(actions, suffix=f'wait{k}.acts') def run(self, k): # R/W actions generated for the whole test set actions = [] # Final translations translations = [] start = time.time() for batch in progress_bar(self.data_loader, unit='batch'): self.clear_states() batch.device(DEVICE) # Compute all at once self.cache_encoder_states(batch) # Read some words and get trimmed states state_dict = self.read_more(k) self.update_s(k) # Initial state is None i.e. 0. state_dict is not used self.prev_h = self.decoder_init(state_dict) # last batch could be smaller than the requested batch size cur_batch_size = batch.size # Start all sentences with <s> self.set_first_word_to_bos(cur_batch_size) # We will start by writing next_action = self.ACT_WRITE while not self.eos_written and self.t_ptr < self.max_len: if next_action == self.ACT_WRITE or self.is_src_read(): logp, new_h, new_word = self.decoder_step( state_dict, self.prev_word, self.prev_h, self.tf_decoder_input) self.write(new_word, new_h) else: # READ state_dict = self.read_more(1) self.update_s(1) # Invert the last committed action for interleaved decoding next_action = 1 - self.actions[-1] # All finished, convert translations to python lists on CPU idxs = self.buffer[self.buffer.ne(0)].tolist() if idxs[-1] != self.eos: # In cases where <eos> not produced and the above loop # went on until max_len, add an explicit <eos> for correctness idxs.append(self.eos) # compute action sequence from which metrics will be computed actions.append(' '.join(map(lambda i: str(i), self.actions))) translations.append(self.vocab.idxs_to_sent(idxs)) up_time = time.time() - start hyps = sort_predictions(self.data_loader, translations) actions = sort_predictions(self.data_loader, actions) return (hyps, actions, up_time)Ancestors
Methods
def run(self, k)-
Expand source code
def run(self, k): # R/W actions generated for the whole test set actions = [] # Final translations translations = [] start = time.time() for batch in progress_bar(self.data_loader, unit='batch'): self.clear_states() batch.device(DEVICE) # Compute all at once self.cache_encoder_states(batch) # Read some words and get trimmed states state_dict = self.read_more(k) self.update_s(k) # Initial state is None i.e. 0. state_dict is not used self.prev_h = self.decoder_init(state_dict) # last batch could be smaller than the requested batch size cur_batch_size = batch.size # Start all sentences with <s> self.set_first_word_to_bos(cur_batch_size) # We will start by writing next_action = self.ACT_WRITE while not self.eos_written and self.t_ptr < self.max_len: if next_action == self.ACT_WRITE or self.is_src_read(): logp, new_h, new_word = self.decoder_step( state_dict, self.prev_word, self.prev_h, self.tf_decoder_input) self.write(new_word, new_h) else: # READ state_dict = self.read_more(1) self.update_s(1) # Invert the last committed action for interleaved decoding next_action = 1 - self.actions[-1] # All finished, convert translations to python lists on CPU idxs = self.buffer[self.buffer.ne(0)].tolist() if idxs[-1] != self.eos: # In cases where <eos> not produced and the above loop # went on until max_len, add an explicit <eos> for correctness idxs.append(self.eos) # compute action sequence from which metrics will be computed actions.append(' '.join(map(lambda i: str(i), self.actions))) translations.append(self.vocab.idxs_to_sent(idxs)) up_time = time.time() - start hyps = sort_predictions(self.data_loader, translations) actions = sort_predictions(self.data_loader, actions) return (hyps, actions, up_time)
Inherited members