95行代码实现最大熵模型训练

关于最大熵模型的介绍请看：http://www.cnblogs.com/hexinuaa/p/3353479.html

下面是GIS训练算法的python实现，代码不到100行。

from collections import defaultdict

import math

class MaxEnt(object):

    def __init__(self):

        self.feats = defaultdict(int)

        self.trainset = []

        self.labels = set()  

      

    def load_data(self,file):

        for line in open(file):

            fields = line.strip().split()

            # at least two columns

            if len(fields) < 2: continue

            # the first column is label

            label = fields[0]

            self.labels.add(label)

            for f in set(fields[1:]):

                # (label,f) tuple is feature 

                self.feats[(label,f)] += 1

            self.trainset.append(fields)

            

    def _initparams(self):

        self.size = len(self.trainset)

        # M param for GIS training algorithm

        self.M = max([len(record)-1 for record in self.trainset])

        self.ep_ = [0.0]*len(self.feats)

        for i,f in enumerate(self.feats):

            # calculate feature expectation on empirical distribution

            self.ep_[i] = float(self.feats[f])/float(self.size)

            # each feature function correspond to id

            self.feats[f] = i

        # init weight for each feature

        self.w = [0.0]*len(self.feats)

        self.lastw = self.w

        

    def probwgt(self,features,label):

        wgt = 0.0

        for f in features:

            if (label,f) in self.feats:

                wgt += self.w[self.feats[(label,f)]]

        return math.exp(wgt)

            

    """

    calculate feature expectation on model distribution

    """        

    def Ep(self):

        ep = [0.0]*len(self.feats)

        for record in self.trainset:

            features = record[1:]

            # calculate p(y|x)