MPC多方安全计算（成功）decisionTree实现

代码地址：csiro-mlai/decision-tree-mpc (github.com) （先运行他给的adult示例代码）【ubuntu 运行环境】

修改成自己的数据集然后进行运行：

按照dockerfile 文件进行配置环境

进入文件夹decision-tree-mpc/：

修改 download.sh 将数据集换成自己的数据集并运行./download.sh （这一步将会把数据集下载到decision-tree-mpc/文件夹下面）【对应的需要修改prepare.py文件，详见下】
安装MP-SPDZ文件：从这个地方下载 zip文件：Releases · data61/MP-SPDZ (github.com)
- cd MP-SPDZ 然后执行 Scripts/tldr.sh （生成所需要的.x文件）
cd MP-SPDZ; make boost mpir
cd .. 然后 ./build-mp-spdz.sh （这里如果git出错参考(57条消息) Git Clone错误解决：GnuTLS recv error (-110): The TLS connection was non-properly terminated._欧晨eli的博客-CSDN博客_gnutls recv error）多尝试几次就好了....
./convert.sh
./run-local.sh emul 2 2 运行即可

预测结果的表示：（计算正确率）

prepare.py文件对应的是对下载到decision-tree-mpc/文件夹下面数据的处理方式（此处要求：先读入所有的label（只能是0/1）,然后读入属性，相当于一列一列的读取所有的数据行：样本个数列：属性+label）【以下代码是对adult数据处理的方式的解读】

adult数据集如下：

0，1，2，3，4，5，6，7，8，9，10，11，12，13，14
39, State-gov, 77516, Bachelors, 13, Never-married, Adm-clerical, Not-in-family, White, Male, 2174, 0, 40, United-States, <=50K
50, Self-emp-not-inc, 83311, Bachelors, 13, Married-civ-spouse, Exec-managerial, Husband, White, Male, 0, 0, 13, United-States, <=50K
38, Private, 215646, HS-grad, 9, Divorced, Handlers-cleaners, Not-in-family, White, Male, 0, 0, 40, United-States, <=50K
53, Private, 234721, 11th, 7, Married-civ-spouse, Handlers-cleaners, Husband, Black, Male, 0, 0, 40, United-States, <=50K
28, Private, 338409, Bachelors, 13, Married-civ-spouse, Prof-specialty, Wife, Black, Female, 0, 0, 40, Cuba, <=50K
37, Private, 284582, Masters, 14, Married-civ-spouse, Exec-managerial, Wife, White, Female, 0, 0, 40, United-States, <=50K
49, Private, 160187, 9th, 5, Married-spouse-absent, Other-service, Not-in-family, Black, Female, 0, 0, 16, Jamaica, <=50K
52, Self-emp-not-inc, 209642, HS-grad, 9, Married-civ-spouse, Exec-managerial, Husband, White, Male, 0, 0, 45, United-States, >50K

#!/usr/bin/python3

import sys

binary = 'binary' in sys.argv
mixed = 'mixed' in sys.argv
nocap = 'nocap' in sys.argv

if binary:
    out = open('binary', 'w')
elif mixed:
    out = open('mixed', 'w')
elif nocap:
    out = open('nocap', 'w')
else:
    out = open('data', 'w')

for start, suffix in (0, 'data'), (1, 'test'):# 这里是从./decision-tree-mpc目录下读取训练样本和测试样本，可以不通过download.sh 来下载，可以直接将数据集放到这个目录下就可以了 （这里必须是两个集合 一个读入train 一个读入test 数据）
    data = [l.strip().split(', ') for l in open('adult.%s' % suffix)][start:-1]

    print(' '.join(str(int(x[-1].startswith('>50K'))) for x in data), file=out) #处理adult的label信息 将 '>50K' 的变成1 其余的label 就是0 （只能处理二分类问题）

    total = 0 #代表了总共的attribute属性的个数
    max_value = 0

    if not binary:
        if nocap:
            attrs = 0, 4, 12
        else:#我们将会进入到这个循环，这里是处理属性是连续变量的数据 （adult中0，2，4，10....列都是连续属性）
            attrs = 0, 2, 4, 10, 11, 12
        for i in attrs:
            print(' '.join(x[i] for x in data), file=out)
            total += 1
            for x in data:
                max_value = max(int(x[i]), max_value)

    if binary or mixed or nocap:#会进入这个循环
        values = [set() for x in data[0][:-1]]
        for x in data:
            for i, value in enumerate(x[:-1]):
                values[i].add(value)
        for i in 1, 3, 5, 6, 7, 8, 9: #对应离散值属性的处理。将其变成one-hot形式的属性来表示
            x = sorted(values[i])
            print('Using attribute %d:' % i,
                  ' '.join('%d:%s' % (total + j, y)
                           for j, y in enumerate(x)))
            total += len(x)
            for y in x:
                print(' '.join(str(int(sample[i] == y)) for sample in data),
                      file=out)

    print(len(data), 'items')
    print(total, 'attributes')
    print('max value', max_value)

如果对应修改成iris.data ,prepare.py 文件如下：

#!/usr/bin/python3

import sys

binary = 'binary' in sys.argv
mixed = 'mixed' in sys.argv
nocap = 'nocap' in sys.argv

if binary:
    out = open('binary', 'w')
elif mixed:
    out = open('mixed', 'w')
elif nocap:
    out = open('nocap', 'w')
else:
    out = open('data', 'w')


for start, suffix in [(0, 'data'),(0,'test')]:
    data = [l.strip().split(',') for l in open('iris.%s' % suffix)]print(' '.join(str(int(x[-1].startswith('Iris-setosa'))) for x in data), file=out)

    total = 0
    max_value = 0

    if not binary:
        if nocap:
            attrs = 0, 4, 12
        else:
            attrs = 0,1, 2,3  #int 类型的数据不需要处理的数据
        for i in attrs:
            print(' '.join(str(int(float(x[i])*100)) for x in data), file=out)
            print(' '.join(str(int(float(x[i])*100)) for x in data))
            total += 1
            for x in data:
                max_value = max(int(float(x[i])), max_value)


    print(len(data), 'items')
    print(total, 'attributes')
    print('max value', max_value)

修改adult.mpc 文件（这个是运行生成决策树的文件）文件位置如下：

对应adult的代码分析：

m = 6 #属性个数
n_train = 32561 #训练集大小
n_test = 16281 #测试集大小

combo = 'combo' in program.args
binary = 'binary' in program.args
mixed = 'mixed' in program.args
nocap = 'nocap' in program.args

try:
   n_threads = int(program.args[2])
except:
   n_threads = None

if combo:
   n_train += n_test

if binary:
   m = 60
   attr_lengths = [1] * m
elif mixed or nocap: #进入这个if
   cont = 6 if mixed else 3 #con 代表连续属性的个数
   m = 60 + cont #二进制（不用管）
   attr_lengths = [0] * cont + [1] * 60 # 0：连续属性 1：离散属性个数（one-hot之后）
else:
   attr_lengths = None

program.set_bit_length(32)
program.options_from_args()

train = sint.Array(n_train), sint.Matrix(m, n_train)
test = sint.Array(n_test), sint.Matrix(m, n_test)

for x in train + test:
    x.input_from(0)

import decision_tree, util

#decision_tree.debug_layers = True
decision_tree.max_leaves = 3000

if 'nearest' in program.args:
   sfix.round_nearest = True

sfix.set_precision_from_args(program, True)

trainer = decision_tree.TreeTrainer(
   train[1], train[0], int(program.args[1]), attr_lengths=attr_lengths,
   n_threads=n_threads)
trainer.debug_selection = 'debug_selection' in program.args
trainer.debug_gini = True
layers = trainer.train_with_testing(*test)

#decision_tree.output_decision_tree(layers)

对应iris 的代码分析：

m = 4 #总共属性的个数
n_train = 124 #训练集个数
n_test = 25 #测试数据的个数

combo = 'combo' in program.args
binary = 'binary' in program.args
mixed = 'mixed' in program.args #采用的这个
nocap = 'nocap' in program.args

try:
   n_threads = int(program.args[2])
except:
   n_threads = None

if combo:
   n_train += n_test

if binary:
   m = 4
elif mixed or nocap:
   cont = 4 #代表连续属性的个数
   m = 4
   attr_lengths = [0] * cont # 
else:
   attr_lengths = None

program.set_bit_length(32)
program.options_from_args()

train = sint.Array(n_train), sint.Matrix(m, n_train)
test = sint.Array(n_test), sint.Matrix(m, n_test)

for x in train + test:
    x.input_from(0)

import decision_tree, util

#decision_tree.debug_layers = True
decision_tree.max_leaves = 3000

if 'nearest' in program.args:
   sfix.round_nearest = True

sfix.set_precision_from_args(program, True)

trainer = decision_tree.TreeTrainer(
   train[1], train[0], int(program.args[1]), attr_lengths=attr_lengths,
   n_threads=n_threads)
trainer.debug_selection = 'debug_selection' in program.args
trainer.debug_gini = True
layers = trainer.train_with_testing(*test)

#decision_tree.output_decision_tree(layers)

过程中如果遇到问题，可以先看下我和这个作者的对话（github issue）：Change to the iris dataset · Issue #2 · csiro-mlai/decision-tree-mpc (github.com)

posted on 2022-12-19 23:24 壳壳· 阅读(220) 评论(0) 编辑收藏举报

会员力量，点亮园子希望

刷新页面返回顶部

导航

MPC多方安全计算（成功）decisionTree实现