def setFeatureWeights(cls, tot_lm_feats, tot_tm_feats, how_many_tm_feats):
assert (tot_lm_feats == len(cls.lmWgt)), "Error: Language model param should have %d weights instead of %d" % (tot_lm_feats, len(cls.lmWgt))
assert (tot_tm_feats == len(cls.tmWgt)), "Error: Translation model param should have %d weights instead of %d" % (tot_tm_feats, len(cls.tmWgt))
for tmVec in cls.tmWgt:
assert (how_many_tm_feats == len(tmVec)), "Error: # of TM features (%d) doesn't match TM weights count (%d)" % (how_many_tm_feats, len(tmVec))
cls.egivenf_offset = 2
StatefulFeatures.setLMInitLst(tot_lm_feats)
cls.setUnkRule()
cls.printWeights()
LanguageModelManager.setLMInfo(cls.lmWgt)
def helperScore(self, newConsItems, is_last_cell):
'''Helper function for computing stateful scores (recomputes lm_heu)'''
(frag_lm_score, lm_comp_heu) = lmm.helperLM(newConsItems, is_last_cell,
self.lmFVec)
self.comp_score += frag_lm_score
self.lm_heu = lm_comp_heu
return frag_lm_score + self.lm_heu
def mergeEntries(self, entriesLst, cube_indx):
# First process the goal: this will be a (regular/glue) rule
sf_f_obj = sff.initNew(entriesLst[0].lm_heu)
score = entriesLst[0].getScoreSansLmHeu()
# Now process the antecedents
anteHyps = []
anteSfFeats = []
anteItemsStates = []
for ante_ent in entriesLst[1:]:
score += ante_ent.getScoreSansLmHeu()
anteHyps.append(ante_ent.tgt)
anteSfFeats.append(ante_ent.sf_feat)
anteItemsStates.append(ante_ent.consItems)
(tgt_hyp, newConsItems) = lmm.helperConsItem(
Lazy.is_last_cell, Lazy.cell_type, Lazy.cell_span, entriesLst[0].tgt.split(), anteHyps, anteItemsStates
)
if settings.opts.force_decode and not Lazy.candMatchesRef(tgt_hyp):
return (score, None) # Hypothesis wouldn't lead to reference; ignore this
"""
Get hypothesis status from the classmethod (in Lazy); hypothesis status can take one of these three values:
-2 : Hyp was not see earlier; create a new entry
-1 : Hyp was seen earlier but current one has a better score; create a new entry to replace the existing one
0 : Hyp was seen earlier and has a poor score than the existing one; ignore this
"""
score_wo_LM = score - sf_f_obj.aggregSFScore(anteSfFeats)
hyp_status = Lazy.getHypothesisStatus(tgt_hyp, score_wo_LM)
""" Should we recombine hypothesis?
A new hypothesis is always added; query LM for lm-score and create new entry_obj.
If an identical hypothesis exists then the current hyp is added under below conditions:
i) the use_unique_nbest flag is False (add new hyp; but use the LM score of the existing one)
ii) use_unique_nbest is True and the new hyp is better than the existing one.
"""
if hyp_status == 0 and settings.opts.use_unique_nbest:
entry_obj = None
else:
score += sf_f_obj.helperScore(newConsItems, Lazy.is_last_cell)
entry_obj = Hypothesis(
score,
self.src_side,
tgt_hyp,
sf_f_obj,
self.depth_hier,
Lazy.cell_span,
entriesLst[0],
entriesLst[1:],
newConsItems,
)
return (score, entry_obj)
def setFeatureWeights(cls, tot_lm_feats, tot_tm_feats, how_many_tm_feats):
assert (
tot_lm_feats == len(cls.lmWgt)
), "Error: Language model param should have %d weights instead of %d" % (
tot_lm_feats, len(cls.lmWgt))
assert (
tot_tm_feats == len(cls.tmWgt)
), "Error: Translation model param should have %d weights instead of %d" % (
tot_tm_feats, len(cls.tmWgt))
for tmVec in cls.tmWgt:
assert (
how_many_tm_feats == len(tmVec)
), "Error: # of TM features (%d) doesn't match TM weights count (%d)" % (
how_many_tm_feats, len(tmVec))
cls.egivenf_offset = 2
StatefulFeatures.setLMInitLst(tot_lm_feats)
cls.setUnkRule()
cls.printWeights()
LanguageModelManager.setLMInfo(cls.lmWgt)
def mergeEntries(self, entriesLst, cube_indx):
# First process the goal: this will be a (regular/glue) rule
sf_f_obj = sff.initNew(entriesLst[0].lm_heu)
score = entriesLst[0].getScoreSansLmHeu()
# Now process the antecedents
anteHyps = []
anteSfFeats = []
anteItemsStates = []
for ante_ent in entriesLst[1:]:
score += ante_ent.getScoreSansLmHeu()
anteHyps.append(ante_ent.tgt)
anteSfFeats.append(ante_ent.sf_feat)
anteItemsStates.append(ante_ent.consItems)
(tgt_hyp, newConsItems) = lmm.helperConsItem(Lazy.is_last_cell, Lazy.cell_type, \
Lazy.cell_span, entriesLst[0].tgt.split(), anteHyps, anteItemsStates)
if settings.opts.force_decode and not Lazy.candMatchesRef(tgt_hyp):
return (score, None
) # Hypothesis wouldn't lead to reference; ignore this
"""
Get hypothesis status from the classmethod (in Lazy); hypothesis status can take one of these three values:
-2 : Hyp was not see earlier; create a new entry
-1 : Hyp was seen earlier but current one has a better score; create a new entry to replace the existing one
0 : Hyp was seen earlier and has a poor score than the existing one; ignore this
"""
score_wo_LM = score - sf_f_obj.aggregSFScore(anteSfFeats)
hyp_status = Lazy.getHypothesisStatus(tgt_hyp, score_wo_LM)
""" Should we recombine hypothesis?
A new hypothesis is always added; query LM for lm-score and create new entry_obj.
If an identical hypothesis exists then the current hyp is added under below conditions:
i) the use_unique_nbest flag is False (add new hyp; but use the LM score of the existing one)
ii) use_unique_nbest is True and the new hyp is better than the existing one.
"""
if (hyp_status == 0 and settings.opts.use_unique_nbest):
entry_obj = None
else:
score += sf_f_obj.helperScore(newConsItems, Lazy.is_last_cell)
entry_obj = Hypothesis(score, self.src_side, tgt_hyp, sf_f_obj, self.depth_hier, Lazy.cell_span, \
entriesLst[0], entriesLst[1:], newConsItems)
return (score, entry_obj)
def stringifyMembers(self, cand_hyp):
return lmm.adjustUNKLMScore(cand_hyp, self.lmFVec)
def helperScore(self, newConsItems, is_last_cell):
'''Helper function for computing stateful scores (recomputes lm_heu)'''
(frag_lm_score, lm_comp_heu) = lmm.helperLM(newConsItems, is_last_cell, self.lmFVec)
self.comp_score += frag_lm_score
self.lm_heu = lm_comp_heu
return frag_lm_score + self.lm_heu
def args():
import optparse
optparser = optparse.OptionParser(usage="usage: cat input | %prog [options]")
optparser.add_option("", "--debug", dest="debug", default=False, action="store_true", help="Debug flag")
optparser.add_option("", "--config", dest="configFile", type="string", help="Configuration file")
optparser.add_option("", "--one-nt-decode", dest="one_nt_decode", default=False, action="store_true", help="Run decoder in 1NT mode (ignore 2NT rules)")
optparser.add_option("", "--shallow-hiero", dest="shallow_hiero", default=False, action="store_true", help="Flag for shallow decoding")
optparser.add_option("", "--shallow-order", dest="sh_order", default=1, type="int", help="Shallow decoding order")
optparser.add_option("", "--free-glue", dest="free_glue", default=True, action="store_true", help="Glue rules can freely combine any X")
optparser.add_option("", "--index", dest="sentindex", default=0, type="int", help="Sentence index")
optparser.add_option("", "--skip-sents", dest="skip_sents", default=None, type="int", help="Skip sentences (usefel to resume decoding mid-way)")
optparser.add_option("", "--sentperfile", dest="sent_per_file", default=500, type="int", help="Sentences per file")
optparser.add_option("", "--fr-rule-terms", dest="fr_rule_terms", default=5, type="int", help="Terms in French side of Hiero rules")
optparser.add_option("", "--inputfile", dest="inFile", type="string", help="Input data file")
optparser.add_option("", "--outputfile", dest="outFile", type="string", help="Output file")
optparser.add_option("", "--glue-file", dest="glueFile", type="string", help="Glue rules file")
optparser.add_option("", "--ttable-file", dest="ruleFile", type="string", help="SCFG rules file")
optparser.add_option("", "--lmodel-file", dest="lmFile", type="string", help="LM file")
optparser.add_option("", "--use-srilm", dest="use_srilm", default=False, action="store_true", help="Flag for using SRILM")
optparser.add_option("", "--no-lm-state", dest="no_lm_state", default=False, action="store_true", help="Don't use LM state for KENLM")
optparser.add_option("", "--no-dscnt-UNKlm", dest="no_dscnt_UNKlm", default=False, action="store_true", help="Don't discount LM penalty for UNK")
optparser.add_option("", "--no-glue-penalty", dest="no_glue_penalty", default=False, action="store_true", help="Don't penalise glue rules")
optparser.add_option("", "--tm-wgt-cnt", dest="tm_weight_cnt", default=5, type="int", help="# of TM weights")
optparser.add_option("", "--trace-rules", dest="trace_rules", default=0, type="int", help="Trace the rules used in the k-best candidates as specified")
optparser.add_option("", "--force-decode", dest="force_decode", default=False, action="store_true", help="Run the decoder in force decode mode")
optparser.add_option("", "--reffile", dest="refFile", type="string", help="Reference file or prefix for multiple refs (for force decoding)")
optparser.add_option("", "--use-local", dest="local_path", default="None", type="string", help="Local path to copy the models")
optparser.add_option("", "--nbest-extremum", dest="nbest_extremum", default=0, type="int", help="Produce nbest_extremum entries if provided; default full nbest list")
optparser.add_option("", "--lm", dest="weight_lm", default=1.0, type="float", help="Language model weight")
optparser.add_option("", "--tm", dest="weight_tm", type="string", help="Translation model weights as a string")
optparser.add_option("", "--tmf", dest="weight_tmf", default=1.0, type="float", help="Forward trans model weight")
optparser.add_option("", "--tmr", dest="weight_tmr", default=1.0, type="float", help="Reverse trans model weight")
optparser.add_option("", "--lwf", dest="weight_lwf", default=0.5, type="float", help="Forward lexical trans weight")
optparser.add_option("", "--lwr", dest="weight_lwr", default=0.5, type="float", help="Reverse lexical trans weight")
optparser.add_option("", "--pp", dest="weight_pp", default=-1.0, type="float", help="Phrase penalty weight")
optparser.add_option("", "--wp", dest="weight_wp", default=-2.0, type="float", help="Word penalty weight")
optparser.add_option("", "--wg", dest="weight_glue", default=0.0, type="float", help="Glue rule weight")
optparser.add_option("", "--cbp", dest="cbp", default=250, type="int", help="Cube pruning pop limit")
optparser.add_option("", "--cbp-diversity", dest="cbp_diversity", default=0, type="int", help="Stack diversity in Cube pruning")
optparser.add_option("", "--ttl", dest="ttl", default=20, type="int", help="# of translations for each source span")
optparser.add_option("", "--btx", dest="beta_x", default=0.001, type="int", help="Beam threshold for X cells")
optparser.add_option("", "--bts", dest="beta_s", default=0.001, type="int", help="Beam threshold for S cells")
optparser.add_option("", "--eps", dest="eps", default=0.1, type="float", help="Beam search margin")
optparser.add_option("", "--1b", dest="one_best", default=False, action="store_true", help="Just do the best derivation")
optparser.add_option("", "--zmert-nbest", dest="zmert_nbest", default=False, action="store_true", help="N-best list should be in zmert format")
optparser.add_option("", "--ng", dest="n_gram_size", default=3, type="int", help="n-gram size")
global opts, feat
(opts, args) = optparser.parse_args()
# Default flags & thresholds
opts.fr_rule_terms = 5
opts.max_phr_len = 10
opts.nbest_limit = 100
opts.use_unique_nbest = True
opts.nbest_format = True
opts.score_diff_threshold = 0.01
opts.elider = '*__*'
opts.lmTupLst = []
opts.weightLM = []
opts.weightTM = []
if opts.configFile is None:
sys.stderr.write('ERROR: Please specify a Config file. Exiting!!')
sys.exit(1)
if opts.configFile is not None:
loadConfig()
if opts.force_decode and not opts.refFile:
sys.stderr.write("ERROR: Forced decoding requires at least one reference file.\n")
sys.stderr.write(" But, no reference file has been specified. Exiting!!\n\n")
sys.exit(1)
if (not opts.no_lm_state) and opts.use_srilm:
sys.stderr.write("INFO: lm_state and srilm are mutually exclusive; no_lm_state can only be used with KENLM.\n")
sys.stderr.write(" Setting no_lm_state to True and using SRILM\n")
opts.no_lm_state = True
if opts.use_srilm:
sys.stderr.write("WARNING: SRILM wrapper is not included with Kriya and needs to be build separately.\n")
sys.stderr.write(" Falling back to use KenLM wrapper.\n")
sys.stderr.write("** If you would like to use SRILM, comment out/remove the lines: 94-98 in Kriya-Decoder/settings.py **\n")
opts.use_srilm = False
sys.stderr.write( "INFO: Using the N-gram size : %d\n" % (opts.n_gram_size) )
sys.stderr.write( "INFO: Run decoder in 1NT mode : %s\n" % (opts.one_nt_decode) )
sys.stderr.write( "INFO: Use X freely in Glue rules : %s\n" % (opts.free_glue) )
sys.stderr.write( "INFO: # of rule terms in Fr side : %d\n" % (opts.fr_rule_terms) )
sys.stderr.write( "INFO: Generating unique N-best : %s\n" % (opts.use_unique_nbest) )
sys.stderr.write( "INFO: Use state info for KENLM : %s\n" % (not opts.no_lm_state) )
sys.stderr.write( "INFO: Discount LM penalty 4 UNK : %s\n" % (not opts.no_dscnt_UNKlm) )
sys.stderr.write( "INFO: Glue rules penalty applied : %s\n" % (not opts.no_glue_penalty) )
sys.stderr.write( "INFO: Cube pruning diversity : %d\n" % (opts.cbp_diversity) )
sys.stderr.write( "INFO: Force decoding status : %s\n" % (opts.force_decode) )
sys.stderr.write( "INFO: Reference file : %s\n" % (opts.refFile) )
if opts.nbest_extremum > 0:
if opts.nbest_extremum * 2 >= opts.nbest_limit:
opts.nbest_extremum = 20
sys.stderr.write( "INFO: Nbest extremum must be less than half the nbest size. Using default nbest extremum of 20.\n" )
else:
sys.stderr.write( "INFO: Nbest extremum set: will produce top-%d and bottom-%d entries as nbest-list\n" % (opts.nbest_extremum, opts.nbest_extremum) )
# Default weights for different features
feat = Features()
if opts.weight_tm:
feat.tm = map( lambda x: float(x), opts.weight_tm.split(' ') )
else:
feat.tm = [opts.weight_tmf, opts.weight_tmr, opts.weight_lwf, \
opts.weight_lwr, opts.weight_pp]
feat.wp = opts.weight_wp
# Set the nbest_format to 'False' & nbest_limit to '1', if one_best option is set
if opts.one_best:
opts.nbest_format = False
opts.nbest_limit = 1
sys.stderr.write("INFO: one-best option specified. Option nbest-format will be turned off and nbest_limit set to 1.\n")
sys.stderr.write( "INFO: cbp/ Nbest limit : %d/ %d\n" % (opts.cbp, opts.nbest_limit) )
if opts.shallow_hiero: sys.stderr.write( "INFO: Shallow decoding hiero with order : %d...\n" % (opts.sh_order) )
else: sys.stderr.write( "INFO: Shallow decoding hiero turned off; decoding as full hiero ...\n" )
if opts.use_srilm: sys.stderr.write( "INFO: Using SRILM language model wrapper ...\n" )
else: sys.stderr.write( "INFO: Using KenLM language model wrapper ...\n" )
# Initialize the language models
LanguageModelManager.initLMs(len(opts.weightLM), opts.lmTupLst, opts.use_srilm)
# Set weights for the features
FeatureManager.glue_wgt = opts.weight_glue
FeatureManager.wp_wgt = opts.weight_wp
FeatureManager.lmWgt = opts.weightLM[:]
FeatureManager.tmWgt = opts.weightTM[:]
FeatureManager.setFeatureWeights(len(opts.weightLM), len(opts.weightTM), opts.tm_weight_cnt)
if opts.local_path is not 'None':
sys.stderr.write( "About to copy language model locally ...\n" )
copyModels()
def scoreRule(self):
p_score = FeatureManager.scorePTEntry(self.sl_feat)
lm_score = LanguageModelManager.scoreLMFeat(self.tgt)
self.lm_heu = lm_score
self.score = p_score + lm_score
def stringifyMembers(self, cand_hyp):
return lmm.adjustUNKLMScore(cand_hyp, self.lmFVec)
def args():
import optparse
optparser = optparse.OptionParser(
usage="usage: cat input | %prog [options]")
optparser.add_option("",
"--debug",
dest="debug",
default=False,
action="store_true",
help="Debug flag")
optparser.add_option("",
"--config",
dest="configFile",
type="string",
help="Configuration file")
optparser.add_option("",
"--one-nt-decode",
dest="one_nt_decode",
default=False,
action="store_true",
help="Run decoder in 1NT mode (ignore 2NT rules)")
optparser.add_option("",
"--shallow-hiero",
dest="shallow_hiero",
default=False,
action="store_true",
help="Flag for shallow decoding")
optparser.add_option("",
"--shallow-order",
dest="sh_order",
default=1,
type="int",
help="Shallow decoding order")
optparser.add_option("",
"--free-glue",
dest="free_glue",
default=True,
action="store_true",
help="Glue rules can freely combine any X")
optparser.add_option("",
"--index",
dest="sentindex",
default=0,
type="int",
help="Sentence index")
optparser.add_option(
"",
"--skip-sents",
dest="skip_sents",
default=None,
type="int",
help="Skip sentences (usefel to resume decoding mid-way)")
optparser.add_option("",
"--sentperfile",
dest="sent_per_file",
default=500,
type="int",
help="Sentences per file")
optparser.add_option("",
"--fr-rule-terms",
dest="fr_rule_terms",
default=5,
type="int",
help="Terms in French side of Hiero rules")
optparser.add_option("",
"--inputfile",
dest="inFile",
type="string",
help="Input data file")
optparser.add_option("",
"--outputfile",
dest="outFile",
type="string",
help="Output file")
optparser.add_option("",
"--glue-file",
dest="glueFile",
type="string",
help="Glue rules file")
optparser.add_option("",
"--ttable-file",
dest="ruleFile",
type="string",
help="SCFG rules file")
optparser.add_option("",
"--lmodel-file",
dest="lmFile",
type="string",
help="LM file")
optparser.add_option("",
"--use-srilm",
dest="use_srilm",
default=False,
action="store_true",
help="Flag for using SRILM")
optparser.add_option("",
"--no-lm-state",
dest="no_lm_state",
default=False,
action="store_true",
help="Don't use LM state for KENLM")
optparser.add_option("",
"--no-dscnt-UNKlm",
dest="no_dscnt_UNKlm",
default=False,
action="store_true",
help="Don't discount LM penalty for UNK")
optparser.add_option("",
"--no-glue-penalty",
dest="no_glue_penalty",
default=False,
action="store_true",
help="Don't penalise glue rules")
optparser.add_option("",
"--tm-wgt-cnt",
dest="tm_weight_cnt",
default=5,
type="int",
help="# of TM weights")
optparser.add_option(
"",
"--trace-rules",
dest="trace_rules",
default=0,
type="int",
help="Trace the rules used in the k-best candidates as specified")
optparser.add_option("",
"--force-decode",
dest="force_decode",
default=False,
action="store_true",
help="Run the decoder in force decode mode")
optparser.add_option(
"",
"--reffile",
dest="refFile",
type="string",
help="Reference file or prefix for multiple refs (for force decoding)")
optparser.add_option("",
"--use-local",
dest="local_path",
default="None",
type="string",
help="Local path to copy the models")
optparser.add_option(
"",
"--nbest-extremum",
dest="nbest_extremum",
default=0,
type="int",
help=
"Produce nbest_extremum entries if provided; default full nbest list")
optparser.add_option("",
"--lm",
dest="weight_lm",
default=1.0,
type="float",
help="Language model weight")
optparser.add_option("",
"--tm",
dest="weight_tm",
type="string",
help="Translation model weights as a string")
optparser.add_option("",
"--tmf",
dest="weight_tmf",
default=1.0,
type="float",
help="Forward trans model weight")
optparser.add_option("",
"--tmr",
dest="weight_tmr",
default=1.0,
type="float",
help="Reverse trans model weight")
optparser.add_option("",
"--lwf",
dest="weight_lwf",
default=0.5,
type="float",
help="Forward lexical trans weight")
optparser.add_option("",
"--lwr",
dest="weight_lwr",
default=0.5,
type="float",
help="Reverse lexical trans weight")
optparser.add_option("",
"--pp",
dest="weight_pp",
default=-1.0,
type="float",
help="Phrase penalty weight")
optparser.add_option("",
"--wp",
dest="weight_wp",
default=-2.0,
type="float",
help="Word penalty weight")
optparser.add_option("",
"--wg",
dest="weight_glue",
default=0.0,
type="float",
help="Glue rule weight")
optparser.add_option("",
"--cbp",
dest="cbp",
default=250,
type="int",
help="Cube pruning pop limit")
optparser.add_option("",
"--cbp-diversity",
dest="cbp_diversity",
default=0,
type="int",
help="Stack diversity in Cube pruning")
optparser.add_option("",
"--ttl",
dest="ttl",
default=20,
type="int",
help="# of translations for each source span")
optparser.add_option("",
"--btx",
dest="beta_x",
default=0.001,
type="int",
help="Beam threshold for X cells")
optparser.add_option("",
"--bts",
dest="beta_s",
default=0.001,
type="int",
help="Beam threshold for S cells")
optparser.add_option("",
"--eps",
dest="eps",
default=0.1,
type="float",
help="Beam search margin")
optparser.add_option("",
"--1b",
dest="one_best",
default=False,
action="store_true",
help="Just do the best derivation")
optparser.add_option("",
"--zmert-nbest",
dest="zmert_nbest",
default=False,
action="store_true",
help="N-best list should be in zmert format")
optparser.add_option("",
"--ng",
dest="n_gram_size",
default=3,
type="int",
help="n-gram size")
global opts, feat
(opts, args) = optparser.parse_args()
# Default flags & thresholds
opts.fr_rule_terms = 5
opts.max_phr_len = 10
opts.nbest_limit = 100
opts.use_unique_nbest = True
opts.nbest_format = True
opts.score_diff_threshold = 0.01
opts.elider = '*__*'
opts.lmTupLst = []
opts.weightLM = []
opts.weightTM = []
if opts.configFile is None:
sys.stderr.write('ERROR: Please specify a Config file. Exiting!!')
sys.exit(1)
if opts.configFile is not None:
loadConfig()
if opts.force_decode and not opts.refFile:
sys.stderr.write(
"ERROR: Forced decoding requires at least one reference file.\n")
sys.stderr.write(
" But, no reference file has been specified. Exiting!!\n\n")
sys.exit(1)
if (not opts.no_lm_state) and opts.use_srilm:
sys.stderr.write(
"INFO: lm_state and srilm are mutually exclusive; no_lm_state can only be used with KENLM.\n"
)
sys.stderr.write(" Setting no_lm_state to True and using SRILM\n")
opts.no_lm_state = True
if opts.use_srilm:
sys.stderr.write(
"WARNING: SRILM wrapper is not included with Kriya and needs to be build separately.\n"
)
sys.stderr.write(" Falling back to use KenLM wrapper.\n")
sys.stderr.write(
"** If you would like to use SRILM, comment out/remove the lines: 94-98 in Kriya-Decoder/settings.py **\n"
)
opts.use_srilm = False
sys.stderr.write("INFO: Using the N-gram size : %d\n" %
(opts.n_gram_size))
sys.stderr.write("INFO: Run decoder in 1NT mode : %s\n" %
(opts.one_nt_decode))
sys.stderr.write("INFO: Use X freely in Glue rules : %s\n" %
(opts.free_glue))
sys.stderr.write("INFO: # of rule terms in Fr side : %d\n" %
(opts.fr_rule_terms))
sys.stderr.write("INFO: Generating unique N-best : %s\n" %
(opts.use_unique_nbest))
sys.stderr.write("INFO: Use state info for KENLM : %s\n" %
(not opts.no_lm_state))
sys.stderr.write("INFO: Discount LM penalty 4 UNK : %s\n" %
(not opts.no_dscnt_UNKlm))
sys.stderr.write("INFO: Glue rules penalty applied : %s\n" %
(not opts.no_glue_penalty))
sys.stderr.write("INFO: Cube pruning diversity : %d\n" %
(opts.cbp_diversity))
sys.stderr.write("INFO: Force decoding status : %s\n" %
(opts.force_decode))
sys.stderr.write("INFO: Reference file : %s\n" %
(opts.refFile))
if opts.nbest_extremum > 0:
if opts.nbest_extremum * 2 >= opts.nbest_limit:
opts.nbest_extremum = 20
sys.stderr.write(
"INFO: Nbest extremum must be less than half the nbest size. Using default nbest extremum of 20.\n"
)
else:
sys.stderr.write(
"INFO: Nbest extremum set: will produce top-%d and bottom-%d entries as nbest-list\n"
% (opts.nbest_extremum, opts.nbest_extremum))
# Default weights for different features
feat = Features()
if opts.weight_tm:
feat.tm = map(lambda x: float(x), opts.weight_tm.split(' '))
else:
feat.tm = [opts.weight_tmf, opts.weight_tmr, opts.weight_lwf, \
opts.weight_lwr, opts.weight_pp]
feat.wp = opts.weight_wp
# Set the nbest_format to 'False' & nbest_limit to '1', if one_best option is set
if opts.one_best:
opts.nbest_format = False
opts.nbest_limit = 1
sys.stderr.write(
"INFO: one-best option specified. Option nbest-format will be turned off and nbest_limit set to 1.\n"
)
sys.stderr.write("INFO: cbp/ Nbest limit : %d/ %d\n" %
(opts.cbp, opts.nbest_limit))
if opts.shallow_hiero:
sys.stderr.write("INFO: Shallow decoding hiero with order : %d...\n" %
(opts.sh_order))
else:
sys.stderr.write(
"INFO: Shallow decoding hiero turned off; decoding as full hiero ...\n"
)
if opts.use_srilm:
sys.stderr.write("INFO: Using SRILM language model wrapper ...\n")
else:
sys.stderr.write("INFO: Using KenLM language model wrapper ...\n")
# Initialize the language models
LanguageModelManager.initLMs(len(opts.weightLM), opts.lmTupLst,
opts.use_srilm)
# Set weights for the features
FeatureManager.glue_wgt = opts.weight_glue
FeatureManager.wp_wgt = opts.weight_wp
FeatureManager.lmWgt = opts.weightLM[:]
FeatureManager.tmWgt = opts.weightTM[:]
FeatureManager.setFeatureWeights(len(opts.weightLM), len(opts.weightTM),
opts.tm_weight_cnt)
if opts.local_path is not 'None':
sys.stderr.write("About to copy language model locally ...\n")
copyModels()