def __init__(self,
heuristic_scores_file="",
collect_stats_strategy='best'):
"""Creates a new ``StatsHeuristic`` instance. The constructor
initializes the unigram table.
Args:
heuristic_scores_file (string): Path to the unigram scores
which are used if this
predictor estimates future
costs
collect_stats_strategy (string): best, full, or all. Defines
how unigram estimates are
collected for heuristic
"""
super(StatsHeuristic, self).__init__()
if heuristic_scores_file:
self.estimates = FileUnigramTable(heuristic_scores_file)
elif collect_stats_strategy == 'best':
self.estimates = BestStatsUnigramTable()
elif collect_stats_strategy == 'full':
self.estimates = FullStatsUnigramTable()
elif collect_stats_strategy == 'all':
self.estimates = AllStatsUnigramTable()
else:
logging.error("Unknown statistics collection strategy")
def __init__(self,
trg_test_file,
accept_subsets=False,
accept_duplicates=False,
heuristic_scores_file="",
collect_stats_strategy='best',
heuristic_add_consumed=False,
heuristic_add_remaining=True,
diversity_heuristic_factor=-1.0,
equivalence_vocab=-1):
"""Creates a new bag-of-words predictor.
Args:
trg_test_file (string): Path to the plain text file with
the target sentences. Must have the
same number of lines as the number
of source sentences to decode. The
word order in the target sentences
is not relevant for this predictor.
accept_subsets (bool): If true, this predictor permits
EOS even if the bag is not fully
consumed yet
accept_duplicates (bool): If true, counts are not updated
when a word is consumed. This
means that we allow a word in a
bag to appear multiple times
heuristic_scores_file (string): Path to the unigram scores
which are used if this
predictor estimates future
costs
collect_stats_strategy (string): best, full, or all. Defines
how unigram estimates are
collected for heuristic
heuristic_add_consumed (bool): Set to true to add the
difference between actual
partial score and unigram
estimates of consumed words
to the predictor heuristic
heuristic_add_remaining (bool): Set to true to add the sum
of unigram scores of words
remaining in the bag to the
predictor heuristic
diversity_heuristic_factor (float): Factor for diversity
heuristic which
penalizes hypotheses
with the same bag as
full hypos
equivalence_vocab (int): If positive, predictor states are
considered equal if the the
remaining words within that vocab
and OOVs regarding this vocab are
the same. Only relevant when using
hypothesis recombination
"""
super(BagOfWordsPredictor, self).__init__()
with open(trg_test_file) as f:
self.lines = f.read().splitlines()
if heuristic_scores_file:
self.estimates = FileUnigramTable(heuristic_scores_file)
elif collect_stats_strategy == 'best':
self.estimates = BestStatsUnigramTable()
elif collect_stats_strategy == 'full':
self.estimates = FullStatsUnigramTable()
elif collect_stats_strategy == 'all':
self.estimates = AllStatsUnigramTable()
else:
logging.error("Unknown statistics collection strategy")
self.accept_subsets = accept_subsets
self.accept_duplicates = accept_duplicates
self.heuristic_add_consumed = heuristic_add_consumed
self.heuristic_add_remaining = heuristic_add_remaining
self.equivalence_vocab = equivalence_vocab
if accept_duplicates and not accept_subsets:
logging.error("You enabled bow_accept_duplicates but not bow_"
"accept_subsets. Therefore, the bow predictor will "
"never accept end-of-sentence and could cause "
"an infinite loop in the search strategy.")
self.diversity_heuristic_factor = diversity_heuristic_factor
self.diverse_heuristic = (diversity_heuristic_factor > 0.0)