Garnet是基于MP-SPDZ开发的一个实用化安全多方学习框架，支持SecureML协议、预训练模型的微调、XGBoost以及决策树模型的安全训练。
当前Garnet支持Linux 2014以及MACOS High Sierra之后的操作系统版本。
首先要求已经按照本框架的安装说明配置好本框架的运行环境，并确保能够使用Script/ring.sh 运行样例程序(tutorial.mpc)。
为了使用该算法进行训练，用户需要提供统一格式的训练数据并使用框架中所提供的脚本对数据进行处理。 用户首先需要在Data目录下存放csv格式的训练集和测试集。命名格式例如：训练集 IRIS_train.csv 测试集 IRIS_test.csv， 即使用 [数据集名]_train.csv 和 [数据集名]_test.csv 来命名。csv文件无需表头，每一行代表一个样本，最后一列代表标签。需要注意的是，本算法在读入小数时，会自动将小数部分截断，因此如果小数部分的数值有意义，请提前对小数部分乘上一个合适的系数并转化为整数。 在准备好csv格式的数据集后，运行python Script/data_prepare_for_xgboost [数据集名] 从而生成符合框架的数据格式，生成的文件为Player-Data/Input-P0-0。运行该脚本后，控制台会输出训练集所包含的训练样本数，特征数，测试集所包含的样本数，特征数。例如:
Garnet % python ./compile.py xgboost -R 32
...
控制台输出
...
Writing to .../xgboost.sch
Writing to .../xgboost-0.bc
Program requires at most:
3335730 integer inputs from player 0
24960385 integer bits
58387325 integer triples
inf bit inverses
3334980 integer inputs from player 1
71918 virtual machine rounds
在准备好上述数据后，根据需要修改Programs/xgboost.mpc。所需修改的信息为前六行。其中其三行根据上个步骤所输出的信息进行修改。
m = 4 # 特征数
n_train = 135 # 训练样本数量
n_test = 15 # 测试样本数量
h = 4 # 树高
n_estimators = 5 # 树的数量
n_threads = 4 # 最大线程数
配置好上述脚本后，即可运行python ./compile.py xgboost -R 32获得编译后的二进制文件，该过程需要花费几分钟时间，且控制台显示的warning可以忽略。之后运行./Script/ring.sh xgboost 获得最后的运行结果。
Garnet % python ./compile.py xgboost -R 32
...
控制台输出
...
Writing to .../xgboost.sch
Writing to .../xgboost-0.bc
Program requires at most:
3335730 integer inputs from player 0
24960385 integer bits
58387325 integer triples
inf bit inverses
3334980 integer inputs from player 1
71918 virtual machine rounds
Garnet % ./Scripts/ring.sh xgboost
Using security parameter 40
Trying to run 32-bit computation
Training the 0-th tree
training 0-th layer
training 1-th layer
training 2-th layer
training 3-th layer
Training the 1-th tree
training 0-th layer
training 1-th layer
training 2-th layer
training 3-th layer
Training the 2-th tree
training 0-th layer
training 1-th layer
training 2-th layer
training 3-th layer
Training the 3-th tree
training 0-th layer
training 1-th layer
training 2-th layer
training 3-th layer
Training the 4-th tree
training 0-th layer
training 1-th layer
training 2-th layer
training 3-th layer
test for train set
true y = [0, 0, 1, 1, 2, 2, 1, 1, 2, 2, 0, 1, 0, 1, 0, 2, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1, 2, 2, 2, 0, 1, 2, 1, 2, 0, 0, 0, 1, 1, 1, 1, 0, 1, 1, 2, 0, 1, 0, 0, 1, 2, 1, 0, 1, 0, 0, 1, 2, 2, 1, 1, 2, 0, 2, 2, 0, 2, 2, 1, 1, 2, 1, 0, 0, 2, 2, 2, 0, 2, 0, 0, 0, 2, 2, 0, 1, 0, 2, 2, 2, 2, 1, 1, 2, 2, 1, 2, 1, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
pred y = [0, 0, 1, 1, 2, 2, 1, 1, 2, 2, 0, 1, 0, 1, 0, 2, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1, 2, 2, 2, 0, 1, 2, 1, 2, 0, 0, 0, 1, 1, 1, 1, 0, 1, 1, 2, 0, 1, 0, 0, 1, 2, 1, 0, 1, 0, 0, 1, 2, 2, 1, 1, 2, 0, 2, 2, 0, 2, 2, 1, 1, 2, 1, 0, 0, 2, 2, 2, 0, 2, 0, 0, 0, 2, 2, 0, 1, 0, 2, 2, 2, 2, 1, 1, 2, 2, 1, 2, 1, 2, 1, 2, 2, 2, 2, 1, 2, 0, 1, 1, 1, 2, 0, 2, 1, 1, 0, 2, 0, 0, 0, 1, 0, 2, 0, 2, 2, 1, 2, 2, 0, 0, 0, 0, 1, 2]
pred y = [0, 0, 0.959335, 0.963852, 1.94325, 1.94325, 0.970718, 0.963852, 1.94325, 1.91367, 0, 0.963852, 0, 0.948578, 0, 1.93993, 0.963852, 0.0685577, 0, 0.963852, 0.292618, 0, 0.963852, 0.963852, 0.747147, 0.963852, 1.93993, 1.94325, 1.91367, 0, 0.959335, 1.93993, 0.963852, 1.93993, 0, 0.163681, 0, 0.963852, 0.962799, 0.963852, 0.963852, 0.0275421, 0.959335, 0.963852, 1.93335, 0, 0.92366, 0, 0, 0.941376, 1.93335, 0.963852, 0, 0.747147, 0, 0.0275421, 0.963852, 1.94325, 1.93993, 0.963852, 0.963852, 1.93993, 0.0977631, 1.93993, 1.9034, 0, 1.93993, 1.93993, 0.959335, 0.963852, 1.93335, 0.963852, 0, 0, 1.93993, 1.91313, 1.93993, 0, 1.94325, 0, 0, -0.0609131, 1.91646, 1.93993, 0.0977631, 0.963852, 0, 1.93993, 1.93335, 1.93993, 1.93993, 0.963852, 0.959335, 1.91492, 1.90637, 0.920074, 1.94022, 0.963852, 1.93993, 0.803986, 1.94325, 1.9097, 1.93993, 1.91678, 0.963852, 1.93993, 0, 0.970718, 0.94223, 0.963852, 1.94325, 0, 1.93993, 0.963852, 0.963852, 0, 1.9034, 0, 0, 0, 0.959824, 0, 1.94325, 0, 1.93993, 1.94325, 0.963852, 1.93993, 1.94325, 0, 0.194946, 0, 0, 0.963852, 1.93993] (not round)
accuracy: 135/135
test for test set
true y = [0, 0, 0, 0, 0, 1, 2, 0, 0, 0, 1, 1, 1, 2, 1]
pred y = [0, 0, 0, 0, 0, 1, 2, 0, 0, 0, 1, 1, 1, 2, 1]
pred y = [0, 0.0275421, 0, -0.0609131, 0.336411, 0.963852, 1.93993, 0, 0, 0, 0.970718, 0.920074, 0.949097, 1.93993, 1.04521] (not round)
accuracy: 15/15
Significant amount of unused dabits of replicated Z2^128. For more accurate benchmarks, consider reducing the batch size with -b.
Significant amount of unused dabits of replicated Z2^32. For more accurate benchmarks, consider reducing the batch size with -b.
Significant amount of unused dabits of replicated Z2^32. For more accurate benchmarks, consider reducing the batch size with -b.
Significant amount of unused dabits of replicated Z2^32. For more accurate benchmarks, consider reducing the batch size with -b.
Significant amount of unused dabits of replicated Z2^32. For more accurate benchmarks, consider reducing the batch size with -b.
Significant amount of unused dabits of replicated Z2^32. For more accurate benchmarks, consider reducing the batch size with -b.
The following benchmarks are including preprocessing (offline phase).
Time = 9.89406 seconds
Data sent = 680.58 MB in ~100377 rounds (party 0; rounds counted double due to multi-threading)
Global data sent = 2041.73 MB (all parties)
This program might benefit from some protocol options.
Consider adding the following at the beginning of 'xgboost.mpc':
program.use_trunc_pr = True
program.use_split(3)
| 数据集名 | 训练集准确率 | 测试机准确率 | |
|---|---|---|---|
| 1 | Kohkiloyeh | 93.33% | 80% |
| 2 | Diagnosis | 100% | 100% |
| 3 | IRIS | 100% | 100% |
| 4 | Wine | 100% | 94.44% |
| 5 | Cancer | 100% | 94.64% |
| 6 | Tic-tac-toe | 90.95% | 88.42% |