analyzer.py 文件源码

def init(args):
    def parse(line):
        attr, pos_id = line.split()
        attr = tuple(attr.split(','))
        return (attr, int(pos_id))
    mappings = Attribute(
        util.OneToOneMapping(
            parse(line) for line in args.pos_def
        ),
        util.OneToOneMapping(
            (row[1], int(row[0])) for row in csv.reader(args.conj_type_def)
        ),
        util.OneToOneMapping(
            (row[1], int(row[0])) for row in csv.reader(args.conj_form_def)
        )
    )
    model = md.Analyzer(
        md.BidirectionalRecognizer(
            md.Recognizer(256, 256, 256, 256),
            md.Recognizer(256, 256, 256, 64 + 256 + 128 + 128)
        ),
        md.Tagger(
            md.BiClassifier(64),
            chainer.ChainList(
                md.Classifier(256, len(mappings.pos)),
                md.Classifier(128, len(mappings.conj_type)),
                md.Classifier(128, len(mappings.conj_form))
            )
        )
    )        
    optimizer = optimizers.AdaGrad(lr=0.01)
    optimizer.setup(model)
    return Storage(mappings, model, optimizer)