Setup

Environment configuration and installation

The tutorial/practical will be done using Python.

See the dedicated official Fidle page for instructions about installing the required environment. More details about instructions for your OS:

• Linux
• Windows
• MacOS

Cloud computing

To simplify the configuration of your environment (and possibly to have access to a GPU¹ for computations during tutorial/practical sessions), we have access to various cloud computing platforms:

• IFB² cloud:
  • connection with institutional account (you should ask to join the “LBMC” group);
  • you can run the Jupyter appliance to get a Jupyter server running Python.
• INSEE³ “datalab SSP cloud”:
  • connection with institutional account through “AgentConnect” combined with “Edugained federation”;
  • you can run the Jupyter-pytorch-gpu service to get a Jupyter server running Python pre-configured with PyTorch and with access to a GPU.

Configuration example on Linux/MacOS

If you are using uv to manage your Python environment, you will find here a uv configuration implementing a working GPU-enabled fidle environment (tested on Linux).

1. Install Python

2. Create a dedicated folder on your computer (change the path to your convenience):

mkdir ~/fidle-tp
cd ~/fidle-tp

3. Create a Python environment for the training:

python3 -m venv fidle-env
source ./fidle-env/bin/activate

When you open a new terminal/session, you will need to run source ./fidle-env/bin/activate once before using any commands/notebooks to activate the related Python environments.

4. Install Pytorch (see the official install helper, especially if you are have a GPU⁴ or not⁵)

# source ./fidle-env/bin/activate # only if not done yet
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

for Mac arm64 you can use the following command:

pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu

5. Install other requirements:

# source ./fidle-env/bin/activate # only if not done yet
pip install torch-geometric torchtext torchdata lightning \
    tensorboard keras transformers numpy Scikit-image Scikit-learn \
    Matplotlib plotly seaborn barviz pyarrow Pandas Pandoc \
    pyyaml Jupyterlab fidle

6. Install notebooks and datasets (using the fid command from the fidle package that we just installed):

# source ./fidle-env/bin/activate # only if not done yet
fid install --quiet

7. Check your setup:

# source ./fidle-env/bin/activate # only if not done yet
fid check

8. Start Jupyter lab:

# source ./fidle-env/bin/activate # only if not done yet
jupyter lab

And you are ready to play!

9. Reinstall and updates

• You can reinstall notebooks or datasets with commands fid install_notebooks and fid install_datasets.
• You can also modify the datasets fidle folder, you need to specify the path using the environment variable FIDLE_DATASETS_DIR.

Footnotes

1. which is not mandatory for the training but can be useful to speed up computations.

2. French Bioinformatics Institute

3. French National Institute of Statistics and Economic Studies

4. You will need to install Nvidia drivers and CUDA library, talk to us if you need any help.

5. Select “CPU” in that case.