help run experiments on tensorflow

raw explanation can be found: https://github.com/panyx0718/models/tree/master/differential_privacy/dp_sgd

下载https://github.com/panyx0718/models 的所有文件存在本地，我的是存在/Users/wanli/Documents/github_code/tensorflow_models/

requirements:

1.Tensorflow 0.10.0 (master branch)

2.Bazel 0.3.1

https://github.com/bazelbuild/bazel/releases?after=0.3.2

下载对应系统的 Bazel0.3.1的sh文件后，确认完成下载后： chmod +x …sh #增加权限 ./..sh #运行安装

3.Download MNIST data for tf

需要的是 mnist_test.tfrecord mnist_train.tfrecord 的格式

如果没有的话， 我是用这个工具下的： https://github.com/panyx0718/models/tree/master/slim

clone／download链接里的所有程序，先下载MNIST源数据，然后使用 convert_mmist.sh完成数据的转换, (更改至相关路径)： 需要开启tensorflow，我好像是pip安装的，所以activate是通过source ~/tensorflow/bin/activate

#!/bin/bash source ~/tensorflow/bin/activate python download_and_convert_data.py –dataset_name=mnist –dataset_dir=”/Users/wanli/Downloads/MNIST/”

至此可以得到文件和相关路径应为以下：

List the codes.

ls -R differential_privacy/ differential_privacy/: dp_sgd init.py privacy_accountant README.md

differential_privacy/dp_sgd: dp_mnist dp_optimizer per_example_gradients README.md

differential_privacy/dp_sgd/dp_mnist: BUILD dp_mnist.py

differential_privacy/dp_sgd/dp_optimizer: BUILD dp_optimizer.py dp_pca.py sanitizer.py utils.py

differential_privacy/dp_sgd/per_example_gradients: BUILD per_example_gradients.py

differential_privacy/privacy_accountant: python tf

differential_privacy/privacy_accountant/python: BUILD gaussian_moments.py

differential_privacy/privacy_accountant/tf: accountant.py accountant_test.py BUILD

List the data.

ls -R data/

./data: mnist_test.tfrecord mnist_train.tfrecord

使用附件里的dp_mnist.py替换/Users/wanli/Documents/github_code/tensorflow_models/differential_privacy/dp_sgd/dp_mnist/dp_mnist.py,有轻微的改动. 整个过程测试并正常运行在我的mac上，但是速度有点慢，为了快速，更改了device参数如下注，没有测试过。

注：

如果无法用gpu运行，相关语句更改为line268: in the tf dp_mnist.py file setting line 268: with tf.Graph().as_default(), tf.Session() as sess, tf.device(‘/cpu:0’): to with tf.Graph().as_default(), tf.Session() as sess, tf.device(‘/gpu:0’,’/gpu:1’):

4.build

#Build （需要在tensorflow_models这一层） #我的pwd: #/Users/wanli/Documents/github_code/tensorflow_models/differential_privacy/dp_sgd/dp_mnist/dp_mnist.py #—————– bazel build -c opt differential_privacy/…

5.run #build完成后不会报错会生成对应的文件，然后运行, 参数格式可能出错，请不要更改

bazel-bin/differential_privacy/dp_sgd/dp_mnist/dp_mnist
–training_data_path=data/mnist_train.tfrecord
–eval_data_path=data/mnist_test.tfrecord
–save_path=tmp/mnist_dir
–num_training_steps=60000
–accountant_type=Moments
–eps=5.0
–projection_dimensions=0

#会开始运行并产生中间结果 … step: 1 step: 2 … step: 9 spent privacy: eps 0.1250 delta 0.72709 spent privacy: eps 0.2500 delta 0.24708 spent privacy: eps 0.5000 delta 0.0029139 spent privacy: eps 1.0000 delta 6.494e-10 spent privacy: eps 2.0000 delta 8.2242e-24 spent privacy: eps 4.0000 delta 1.319e-51 spent privacy: eps 8.0000 delta 3.3927e-107 train_accuracy: 0.53 eval_accuracy: 0.53 …

#运行完成后会 ls tmp/mnist_dir/ checkpoint ckpt ckpt.meta results-0.json