SSDS 2017 - 2^nd Int'l Summer School on Data Science

Center of Research Excellence for Data Science and Advanced Cooperative Systems, Research Unit for Data Science

Hands-on sessions for 2nd International Summer School on Data Science organized by the Center of Research Excellence for Data Science and Advanced Cooperative Systems, Research Unit for Data Science, from September 25-29, 2017 in Split, Croatia.

Table Of Contents

• Day 1 - Introduction to Tensorflow
• Day 2 - Convolutional Neural Networks for Image Classification
• Day 3 - Character-wise Language Modeling with LSTMs
• Day 4 - Image Segmentation and Object Detection
• Day 5 - Generative Adversarial Networks

Installation instructions

The hands-on sessions are organized as Jupyter notebooks, which you can run on your local computer. They were tested on Windows, Linux and Mac OS X within Anaconda (Python 3.6) environment (Version 4.4.0) with TensorFlow (Version 1.3). It is adequate and recommended to have installed TensorFlow with CPU-support only, since it is much easier to install.

Please follow below the specific instructions for your OS (Windows, Linux, Mac OS X) to:

• Download & install Anaconda (from https://www.anaconda.com/download/)
• Create conda environment with Python 3.6 and standard Anaconda Python packages (see conda-cheatsheet)
• Install TensorFlow 1.3 with CPU-support only by using pip (follow https://www.tensorflow.org/install/)

Windows

TensorFlow installation was tested on 64-bit Windows 7 and Windows 10 (there is currently no support for TensorFlow on 32-bit systems). We recommend to install TensorFlow in an Anaconda environment. Download Anaconda (Python 3.6) for Windows from https://www.anaconda.com/download/ and install it by running setup in Administrator mode.

Afterwards, run Anaconda Command Prompt in Administrator mode and

• Create conda environment named ssds:

  conda create --name ssds python=3.6 anaconda
  

• Activate ssds with:

  activate ssds
  

• Install Tensorflow within ssds:

  pip install --ignore-installed --upgrade tensorflow
  

For more more details on how to install TensorFlow on Windows go to https://www.tensorflow.org/install/install_windows. If you already have older version of Anaconda installed where Python 3.6 is not available by default you can still follow the procedure above. It will install Python 3.6 and make it available in the ssds environment.

Linux/Mac OS X

Similarly as for Windows, the most convenient way for TensorFlow installation on Linux or Mac OS X is to install TensorFlow with 'CPU-support only within Anaconda environment' following the similar procedure to the above-described steps. For detailed instructions:

• for Linux go to https://www.tensorflow.org/install/install_linux#installing_with_anaconda

• or for Mac OS X go to https://www.tensorflow.org/install/install_mac#installing_with_anaconda

Clone the repository

First make sure you have git versioning software, you can download and install it from https://git-scm.com/downloads. Using the command line (i.e. [Anaconda] Command Prompt on Windows or Terminal on Linux/Mac OS X) position to the directory of your choice and then clone the git repository with command:

git clone https://github.com/SSDS-Croatia/SSDS-2017.git

As an alternative, you can just download the zip file from https://github.com/SSDS-Croatia/SSDS-2017/archive/master.zip, but then you will have to do it all over again if there are any changes in the meantime!

Using hands-on sessions Jupyter notebooks:

All the hands-on sessions are run as Jupyter notebooks. To view and edit notebooks, in the command line (i.e. using Anaconda Command Prompt on Windows or Terminal on Linux/Mac OS X), run the following:

• activate ssds conda environment (consult conda-cheatsheet):

  • Windows: activate ssds

  • Linux/Mac OS X: source activate ssds

• within ssds start Jupyter Notebook server (for details):

  jupyter notebook --notebook-dir=PATH-TO-REPOSITORY --port=NUMBER-OF-UNUSED-PORT
  

  This will open your default web browser at http://localhost:[NUMBER-OF-UNUSED-PORT] and you will be able to navigate the repository file structure and use notebooks for the hands-on sessions.