def weight_corpus(corpus_inst, timbl_out, weight_func):
"""
Weight corpus instances
@param corpus_inst: CorpusInst instance
@param timbl_out: iterator over Timbl output
@param weight_func: weighting fuction
"""
for graph_inst in corpus_inst:
for inst, (inst_str, k_nn_list) in zip(graph_inst, timbl_out):
feats_str, true_class, pred_class, distrib_str, distance = \
parse_inst(inst_str, with_distrib=True, with_distance=True)
inst["pred_relation"] = pred_class
inst["pred_weight"] = weight_func(
category=pred_class,
distribution=parse_distrib(distrib_str))
def classify(self, instances):
"""
adds predicted class and associated weight to instances
@param instances: numpy.ndarray instance
"""
for inst in instances:
# Assumes that last field in instance is the true class
inst_str = "\t".join( self._to_str(value)
for value in inst )
result = self._client.classify(inst_str)
inst["pred_relation"] = result["CATEGORY"]
# The Timbl client is lazy and does not automatically parse the
# distribution string, so we use parse_distrib to obtain an
# iterator over (class, count) pairs
distribution = parse_distrib(result["DISTRIBUTION"])
inst["pred_weight"] = self.weight_func(
category=result["CATEGORY"],
distribution=distribution)