model.py 文件源码

def __init__(self, num_class=2):
        """
        :type num_classes: int
        :rtype: None
        """

        self.__ctrl__ = None
        self.__case__ = None

        with open('../../.dbname', 'r') as f:
            self.__DB_NAME__ = json.load(f)['dbname']
        self.__MG_DOCS_COLL__   = 'raw-docs'           # raw docs
        self.__MG_SENTS_COLL__  = 'bag-of-sents'       # raw sentences
        self.__MG_TOKENS_COLL__ = 'sample-tokens'      # clean tokens (words)
        self.__PG_STATS_TBL__   = 'stats'              # stylometric features
        self.__PG_RESULTS_TBL__ = 'results_' + \
                                  str(num_class) + \
                                  'class'              # cross val results
        self.__PG_PROBAS_TBL__  = 'probabilities'      # cross val probabilities


        self.__model__ = Pipeline([ \
                                 # ('scaler2', StandardScaler()),
                                 # ('scaler', MinMaxScaler()),
                                 # ('scaler3', Normalizer()),
                                  ('classifier', SVC(probability=True,
                                                     kernel='poly',
                                                     degree=2,
                                                     class_weight='balanced') \
                                                 if num_class-1 \
                                            else OneClassSVM(kernel='rbf',
                                                             nu=0.7,
                                                             gamma=1./250))
                                 ])

        print 'Instantiated classifier %s.' % \
              self.__model__.named_steps['classifier'].__class__.__name__


        self.__io__ = DBIO(MG_DB_NAME=self.__DB_NAME__,
                           PG_DB_NAME=self.__DB_NAME__)

        self.__tagger__ = None     # initialise if re-creating samples
        self.__bootstrap__ = None  # initialise in fit