OpenSBT MNIST - Fashion-MNIST

This repository integrates the MNIST and Fashion-MNIST case study into OpenSBT and is based on the implementation in DeepJanus.

Preliminaries

Download Repository

git clone https://github.com/opensbt/opensbt-fmnist.git
cd opensbt-fmnist

Installing Python 3.8

Install Python 3.8

add-apt-repository ppa:deadsnakes/ppa
apt update
apt install -y python3.8

And check if it is correctly installed, by typing the following command:

$ python3.8

You should have a message that tells you are using python 3.8

Installing pip

Use the following commands to install pip and upgrade it to the latest version:

apt install -y python3-pip
python3.8 -m pip install --upgrade pip

Once the installation is complete, verify the installation by checking the pip version:

python3.8 -m pip --version

Creating a Python virtual environment

Install the venv module in the docker container:

apt install -y python3.8-venv

Create the python virtual environment:

python3.8 -m venv .venv

Activate the python virtual environment and updated pip again (venv comes with an old version of the tool):

. .venv/bin/activate
pip install --upgrade pip

Installing OpenSBT Requirements

pip  install -r requirements.txt

Installing Python Binding to the Potrace library

Instructions provided by https://github.com/flupke/pypotrace.

Install system dependencies in your environment (it is not needed to install them in the MNIST folder):

apt-get install build-essential python3,8-dev libagg-dev libpotrace-dev pkg-config 

Install pypotrace (commit 76c76be2458eb2b56fcbd3bec79b1b4077e35d9e):

git clone https://github.com/flupke/pypotrace.git
cd pypotrace
git checkout 76c76be2458eb2b56fcbd3bec79b1b4077e35d9e
pip install numpy
pip install setuptools==65.5.0 wheel Cython==0.29.34
pip install .
cd ..

If the following command does not crash, pypotrace is correctly installed:

python
>>> import potrace
>>>

Installing PyCairo and PyGObject

Instructions provided by https://pygobject.readthedocs.io/en/latest/getting_started.html#ubuntu-getting-started.

Open a terminal and execute

apt-get install python3-gi python3-gi-cairo gir1.2-gtk-3.0

And

apt-get install libgirepository1.0-dev gcc libcairo2-dev pkg-config python3-dev gir1.2-gtk-3.0 librsvg2-dev

Installing MNIST requirements

cd mnist/
pip install -r requirements.txt
cd ...

Problem Definition

The mutation of digits is performed by selection of control points/vertices of the corresponding svg model of the digit and performing a displacment in x and y direction of that point. The index of the point and extent represent the search parameters of the search based testing problem. Right now, two configs are supported

• 3D Problem. (diplacement_1, displacement_2, index_vertex). Displaces vertex with given index and the following vertex with provided displacement.
• 6D Problem. (diplacement_1, displacement_2, displacement_3, diplacement_4, index_vertex_1, index_vertex_2). Displaces vertex_1 and vertex_2 with provided displacement.

There are multiple fitness functions implemented integrating static image properties (num turns, saturation, etc.).

Run example for MNIST

Example OpenSBT experiments can be found in default_experiments.py. To run an experiment via console use syntax from OpenSBT:

python run.py -e 1000

To see an example experiment definition without flag based execution check out run_mnist.py.

Run example for Fashion-MNIST

Example OpenSBT experiments can be found in default_experiments_fmnist.py. To run an experiment via console use syntax from OpenSBT:

python run.py -e 201

To see an example experiment definition without flag based execution check out run_fmnist.py.

Lev Sorokin (lev.sorokin@tum.de)