Installation classique sous Linux (pip)

About this installation

  • This installation procedure is for Windows 10 and more.
  • This installation procedure is based on the officia PyTorch procedure,
    which are described here : https://pytorch.org/
  • If you have a GPU card, things can be more complicated than this simplified procedure.
    We advise you to consult the documentation mentioned above.
  • Use cmd shell, not powershell…
  • Have you tried Docker or Linux or WSL2 ? …you should ;-P
  • Size a full installation (environment, data, notebooks) is around 4GB

Content

Installation

1. First of all, you need Python !

A recent version of python 3.9 to 3.10 is required.
You can get it from : https://www.python.org/downloads/

Note :

  • version 3.11 and more can sometimes cause problems
  • Remember to select the option “add to path”

2 - A folder to contain them all

The idea is to put everything in the same folder, for example fidle-tp
You can create your folder from the explorer or from a terminal.
For example : C:\fidle-tp

3 - A python environment, just for us !

The idea is to create a virtual python environment, in order not to “pollute” your system.
This virtual environment will be contained in a sub folder (fidle-env).

FIRST, we create a virtual environment “fidle-env” and activate it.

To create your virtual env and activate it, get a cmd shell (not powershell !) and run these commands :

C:\> cd C:\fidle-tp
C:\fidle-tp> python -m venv --system-site-packages fidle-env
C:\fidle-tp> fidle-env\Scripts\activate
(fidle-env) C:\fidle-tp>

SECOND, we have to install PyTorch, according PyTorch documentation : https://pytorch.org/get-started/locally/

Example: on Windows for cpu :

pip3 install torch torchvision torchaudio

THIRD, we will now install the rest of the modules:-)

(fidle-env) $ 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

It will take a “little while”, but should go very well :-)

4 - Install notebooks et datasets

Now all that remains is to install the notebooks and datasets :-)
We will do this with Fidle’s fid command.

Still from your fidle-tp folder

(fidle-env) C:\fidle-tp> fid install --quiet

It will take a little while again, but should go very well again:-)

5 - Installation check

The tree structure is normally as follows:

 fidle-tp
     ├── fidle                 Contains notebooks
     ├── fidle-datasets        Contains datasets
     └── fidle-env             Virtual environment

From a terminal, go to your fidle-tp folder,
activate the fidle-env environment and do a fid check.

C:> cd C:\fidle-tp
C:\fidle-tp> fidle-env\Scripts\activate
(fidle-env) C:\fidle-tp> fid check

You should have something like :
(version numbers may vary) (Don’t be too worried about possible warnings…)

==========================================================
fid - Your favorite Fidle admin command :-)    (v2.3.0)
==========================================================

Notebooks and datasets can only be found if they are in/near the explored folder.
Explored directory is : c:\Placard\tmp\fidle-tp

Datasets dir found :

    c:\fidle-tp\datasets-fidle               (Datasets Fidle / 2.0)
    The environment variable FIDLE_DATASETS_DIR is :     undefined

Notebooks dir found :

    c:\fidle-tp\fidle-master                 (Notebooks Fidle / 3.0.5)

Check environment :

    Python               : Ok         (3.10.8)
    Fidle module         : Ok         (2.3.0)
    keras                : Ok         (3.0.4)
    numpy                : Ok         (1.26.3)
    sklearn              : Ok         (1.4.0)
    yaml                 : Ok         (6.0.1)
    skimage              : Ok         (0.22.0)
    matplotlib           : Ok         (3.8.2)
    plotly               : Ok         (5.18.0)
    pandas               : Ok         (2.2.0)
    jupyterlab           : Ok         (4.0.11)
    torch                : Ok         (2.1.2+cpu)
    torchvision          : Ok         (0.16.2+cpu)
    lightning            : Ok         (2.1.3)

6 - Start Jupyter lab

Very easy :-)
From a terminal, go to your fidle-tp folder,
activate the fidle-env environment and do a jupyter lab.

c:> cd C:\fidle-tp
C:\fidle-tp> fidle-env\Scripts\activate
(fidle-env) C:\fidle-tp> jupyter lab

7 - Reinstall and updates

  • You can reinstall notebooks or datasets with the fid install_notebooks and fid install_datasets commands
  • You can also locate the datasets fidle folder anywhere you want.
    In this case, you must specify the location of this folder with the environment variable FIDLE_DATASETS_DIR
Fidle