dpml-benchmark - GitLite

KO | EN

gitlite — search

#python #react #golang #android #rust #deep-learning #docker #java #awesome #kubernetes #vue #awesome-list #ios #open-source

dpml-benchmark

(★ 63)

This repository contains the codes for first large-scale investigation of Differentially Private Convex Optimization algorithms.

파일 탐색기

최종 버전 다운로드 (.zip)

- approximate_minima_perturbation.py
- frank_wolfe.py
- gradient_descent.py
- psgd.py
- __init__.py
- clipping.py
- common.py
- constraints.py
- datasets.py
- gaussian_moments.py
- graph.py
- noise.py
- tuning_driver.py
- - .gitkeep
- - .gitkeep
- - adult.py
  - covertype.py
  - gisette.py
  - kddcup99.py
  - mnist.py
  - rcv1.py
  - realsim.py
- - utils_download.py
  - utils_preprocessing.py
  - utils_sparse.py
- main_preprocess.py
- __init__.py
- huber_svm.py
- logistic_regression.py
.gitignore
draw.py
gridsearch.py
LICENSE
README.md
requirements.txt

# 설치 가이드

1. 코드 내려받기

git clone https://github.com/sunblaze-ucb/dpml-benchmark

깃허브에서 프로젝트 코드 전체를 내 컴퓨터로 내려받습니다.

cd dpml-benchmark

방금 내려받은 프로젝트 폴더 안으로 이동합니다.

2. Python

쉬움 추천

사전 준비물

Git GitHub에서 프로젝트 코드를 내려받으려면 필요합니다.
Python 3 설치 시 'Add Python to PATH' 옵션을 꼭 체크하세요 (Windows).

python3 -m venv venv

파이썬 스크립트(또는 모듈)를 실행합니다.

pip install -r requirements.txt

requirements.txt 등에 명시된 파이썬 라이브러리를 설치합니다.

✅ 에러 메시지 없이 실행되고 터미널에 안내 문구가 출력되면 정상입니다.

이 레포의 README에 적힌 실제 명령어를 그대로 가져왔습니다.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Differentially Private Convex Optimization Benchmark

A benchmark and implementations for differentially private convex optimization algorithms.

The algorithms implemented in this repository are as follows:
1. Approximate Minima Perturbation - An original algorithm proposed in our paper.
2. Hyperparameter-free Approximate Minima Perturbation - A hyperparameter-free version of Approximate Minima Perturbation.
3. Private Stochastic Gradient Descent from [BST'14](https://arxiv.org/abs/1405.7085), [ACTMMTZ'16](https://arxiv.org/pdf/1607.00133.pdf)
4. Private Convex Perturbation-Based Stochastic Gradient Descent from [WLKCJN'17](https://arxiv.org/pdf/1606.04722.pdf)
5. Private Strongly Convex Perturbation-Based Stochastic Gradient Descent from [WLKCJN'17](https://arxiv.org/pdf/1606.04722.pdf)
6. Private Frank-Wolfe from [TTZ'16](https://arxiv.org/pdf/1411.5417.pdf)

## Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes. The codes are currently implemented using NumPy. They require Python version 3.5 or newer. You will also need to install all dependencies listed in the requirements.txt file in the repository. The recommended way to do this is through the use of a Python virtual environment.

### Virtual Environment

You can set up a virtual environment as follows:

1. Navigate to the directory that you have checked out this repository in.
2. Create a virtual environment named *venv* by running:
```bash
python3 -m venv venv
```
If any needed packages are missing, you should get an error message telling you which ones to install.

3. Activate the virtual environment by running the following command on Posix systems:
```bash
source venv/bin/activate
```
There will be a script for Windows systems located at *venv/Scripts/activate*. However, none of the code in this repository has been tested on Windows.

### Prerequisites

```
cycler==0.10.0
matplotlib==2.0.2
numpy==1.13.0
pyparsing==2.2.0
python-dateutil==2.6.0
pytz==2017.2
scipy==0.19.0
scikit-learn==0.18.1
six==1.10.0
xlrd==1.0.0
tensorflow==1.11.0
```

### Installing

#### Linux

1. Navigate to the this repository.
2. Run the following command line.

```bash
pip install -r requirements.txt
```

#### Windows

1. Navigate to the this repository.
2. Open requirements.txt in the repo, and run ''pip install'' for all of the prequisities in order.

## Running the benchmark

### Download and preprocess the datasets

1. Navigate to the ''datasets'' directory.
2. Run the following command line to download and preprocess all the benchmark datasets automatically.
```bash
python main_preprocess.py all
```
3. If you want to download one of the datasets, just replace ''all'' with the name of the dataset. All available datasets are listed as following.
```
adult, covertype, gisette, kddcup99, mnist, realsim, rcv1
```

### Run the benchmarks

1. Navigate to this repository.
2. Run algorithms on one dataset using the following command.

```bash
python gridsearch.py [ALG_NAME] [DATASET_NAME] [MODEL_NAME]
```

3. Available ALG_NAME
```
ALL: all the algorithms
AMP: Approximate Minima Perturbation
AMP-NT: Hyperparameter-free Approximate Minima Perturbation
PSGD: Private Stochastic Gradient Descent
PPSGD: Private Convex Perturbation-Based Stochastic Gradient Descent
PPSSGD: Private Strongly Convex Perturbation-Based Stochastic Gradient Descent
FW: Private Frank-Wolfe
```

4. Available DATASET_NAME
```
adult, covertype, gisette, kddcup99, mnist, realsim, rcv1
```
5. Available MODEL_NAME
```
LR: Logistic Regression
SVM: Huber SVM without kernel functions
```
6. The results are stored in csv files in ''dpml-algorithms/results/rough_results''

### Draw the graphs

1. Navigate to this repository.
2. Run the following command to get the graph after running the corresponding benchmark.
```bash
python draw.py [DATASET_NAME] [ALG_NAME] [MODEL_NAME]
```
3. The graphs are in ''dpml-algorithms/results/graphs''.