def create_decoder(config_str, weights):
"""
Creates an instance of cdec.Decoder
:param config_str: cdec's configuration string
:param dict weights: parameters of the instrumental distribution
:return: an instance of cdec.Decoder
"""
# perhaps make sure formalism=scfg and intersection_strategy=full?
# decoder = cdec.Decoder(config_str=config_str, formalism='scfg', intersection_strategy='Full')
decoder = cdec.Decoder(config_str=config_str)
for k, v in weights.iteritems():
decoder.weights[k] = v
return decoder
def create_decoder_from_files(cdec_ini, weights_file, scaling=1.0):
"""
Creates an instance of cdec.Decoder
:param cdec_ini: cdec's configuration file
:param weights_file: parameters of the instrumental distribution
:param scaling: scaling factor (defaults to 1.0)
:return: an instance of cdec.Decoder
"""
with open(cdec_ini) as f:
config_str = f.read()
logging.info('cdec.ini:\n\t%s', '\n\t'.join(config_str.strip().split('\n')))
# perhaps make sure formalism=scfg and intersection_strategy=full?
# decoder = cdec.Decoder(config_str=config_str, formalism='scfg', intersection_strategy='Full')
decoder = cdec.Decoder(config_str=config_str)
logging.info('Loading weights: %s', weights_file)
decoder.read_weights(weights_file, scaling)
# logging.info('Weights: %s', dict(decoder.weights))
return decoder
def main():
logging.basicConfig(level=logging.INFO, format='%(message)s')
parser = argparse.ArgumentParser()
parser.add_argument('-c', '--config', help='cdec config', required=True)
parser.add_argument('-w', '--weights', help='initial weights', required=True)
parser.add_argument('-r', '--reference', help='reference file', required=True)
parser.add_argument('-s', '--source', help='source file', required=True)
args = parser.parse_args()
with open(args.config) as fp:
config = fp.read()
decoder = cdec.Decoder(config)
decoder.read_weights(args.weights)
with open(args.reference) as fp:
references = fp.readlines()
with open(args.source) as fp:
sources = fp.readlines()
assert len(references) == len(sources)
rampion(decoder, sources, references)
for fname, fval in sorted(dict(decoder.weights).iteritems()):
print('{0}\t{1}'.format(fname, fval))
def setUp(self):
self.decoder = cdec.Decoder(formalism='scfg')
self.decoder.read_weights(weights)
with gzip.open(grammar_file) as f:
self.grammar = f.read()
#coding: utf8
import cdec
import gzip
from math import exp
weights = '../tests/system_tests/australia/weights'
grammar_file = '../tests/system_tests/australia/australia.scfg.gz'
# Load decoder width configuration
decoder = cdec.Decoder(formalism='scfg')
# Read weights
decoder.read_weights(weights)
print dict(decoder.weights)
# Read grammar
with gzip.open(grammar_file) as f:
grammar = f.read()
# Input sentence
sentence = u'澳洲 是 与 北韩 有 邦交 的 少数 国家 之一 。'
print ' Input:', sentence.encode('utf8')
# Decode
forest = decoder.translate(sentence, grammar=grammar)
# Get viterbi translation
print 'Output[0]:', forest.viterbi().encode('utf8')
f_tree, e_tree = forest.viterbi_trees()
print ' FTree[0]:', f_tree.encode('utf8')
print ' ETree[0]:', e_tree.encode('utf8')
def make_decoder(config, weights):
global decoder
decoder = cdec.Decoder(config)
decoder.read_weights(weights)