有监督的卡方分箱算法

0|1实现代码

import numpy as np
import pandas as pd
from collections import Counter
def chimerge(data, attr, label, max_intervals):
    distinct_vals = sorted(set(data[attr])) # Sort the distinct values
    labels = sorted(set(data[label])) # Get all possible labels
    empty_count = {l: 0 for l in labels} # A helper function for padding the Counter()
    intervals = [[distinct_vals[i], distinct_vals[i]] for i in range(len(distinct_vals))] # Initialize the intervals for each attribute
    while len(intervals) > max_intervals: # While loop
        chi = []
        for i in range(len(intervals)-1):
            # Calculate the Chi2 value
            obs0 = data[data[attr].between(intervals[i][0], intervals[i][1])]
            obs1 = data[data[attr].between(intervals[i+1][0], intervals[i+1][1])]
            total = len(obs0) + len(obs1)
            count_0 = np.array([v for i, v in {**empty_count, **Counter(obs0[label])}.items()])
            count_1 = np.array([v for i, v in {**empty_count, **Counter(obs1[label])}.items()])
            count_total = count_0 + count_1
            expected_0 = count_total*sum(count_0)/total
            expected_1 = count_total*sum(count_1)/total
            chi_ = (count_0 - expected_0)**2/expected_0 + (count_1 - expected_1)**2/expected_1
            chi_ = np.nan_to_num(chi_) # Deal with the zero counts
            chi.append(sum(chi_)) # Finally do the summation for Chi2
        min_chi = min(chi) # Find the minimal Chi2 for current iteration
        for i, v in enumerate(chi):
            if v == min_chi:
                min_chi_index = i # Find the index of the interval to be merged
                break
        new_intervals = [] # Prepare for the merged new data array
        skip = False
        done = False
        for i in range(len(intervals)):
            if skip:
                skip = False
                continue
            if i == min_chi_index and not done: # Merge the intervals
                t = intervals[i] + intervals[i+1]
                new_intervals.append([min(t), max(t)])
                skip = True
                done = True
            else:
                new_intervals.append(intervals[i])
        intervals = new_intervals
    for i in intervals:
        print('[', i[0], ',', i[1], ']', sep='')

0|1使用例子

iris = pd.read_csv('http://archive.ics.uci.edu/ml/machine-learning-databases/iris/iris.data', header=None)
iris.columns = ['sepal_l', 'sepal_w', 'petal_l', 'petal_w', 'type']
for attr in ['sepal_l', 'sepal_w', 'petal_l', 'petal_w']:
    print('Interval for', attr)
    chimerge(data=iris, attr=attr, label='type', max_intervals=3)

结果:

__EOF__

作　　者：Hichens
出　　处：https://www.cnblogs.com/hichens/p/13585854.html
关于博主：莫得感情的浅度学习机器人
版权声明：@Hichens
声援博主：如果您觉得文章对您有帮助，可以点击文章右下角【推荐】一下。您的鼓励是博主的最大动力！