Skip to content
Snippets Groups Projects

Install Fidle v1

Last edited by Soraya Arias

NOTE : Cette procédure sera prochainement mise à jour pour la session 2 de Fidle !
NOTE : This procedure will be updated soon for Fidle session 2 !

Important :

  • The procedure below is validated under Linux, remember that Debian is good for you !
  • Under MacOS, the installation is a priori possible manually, but sorry, we won't be able to help you much...
  • Under Windows, some instructions are given, the use on miniconda 3.8 for Windows is helpful, as well as the use of Anaconda Prompt terminal. But only the manual installation of fidle environement has been validated.

Ressources :

  • All notebooks can run on CPU or GPU
  • The software environment (conda) requires about 10G
  • Original datasets require about 2.5G
  • Consolidated datasets may require up to 150G depending on the tests you perform... but 15GB will probably be a minimum...

The installation consists of 3 steps:

1/ Get Fidle repository
2/ Get datasets
3/ Setup Fidle environment
4/ Start Jupyter lab

1/ Get Fidle repository

The simplest and most conventional way is with git (100 MB):

git clone https://gricad-gitlab.univ-grenoble-alpes.fr/talks/fidle.git

However, a classic retrieval (zip, tar, ?...) is possible via the download button, near [Clone] button.

Important : The project is frequently updated and we encourage you to get the latest version. The version number is specified in the README of the repository.

2/ Get Datasets

Get it
The datasets used in the examples are available here.
You can retrieve and decompress them using the following commands:

wget https://cloud.univ-grenoble-alpes.fr/index.php/s/XAHB7PRmTabTTz7/download -O datasets.tar  
tar -xf datasets.tar

Specify the location
You must then indicate the location of this folder with an environment variable.

  • For computers under Linux, add the following line to your .bashrc, or `.bash_profile, ...
export FIDLE_DATASETS_DIR=<the place of your datasets directory>
  • For computers under Windows (and using a terminal that is NOT a powershell!)
set FIDLE_DATASETS_DIR=<the place of your datasets directory>

A complete description of the different datasets is available in Notebooks.

3/ Setup Fidle Environment

Easy way (but may generate conflit errors...)

Fidle Jupyter notebooks need a special but classic deep learning Python environment.

We prepare a conda environment (for Linux OS and for Windows OS) that can easily be installed when using conda. The conda environment file is located at <PATH_TO_CLONED_FIDLE_REPO>/fidle_environment_[linux|windows10].txt.

This environment can be created using conda tool with the following 1-line command (need 6 GB min.):

For Linux

conda create --name fidle --file <PATH_TO_CLONED_FIDLE_REPO>/fidle_environment_linux.txt 
conda create --name fidle --file <PATH_TO_CLONED_FIDLE_REPO>/fidle_environment_linux_gpu_cuda101.txt

For Windows

=> with miniconda 3.8 installed and using a Anaconda prompt terminal

conda create --name fidle --file <PATH_TO_CLONED_FIDLE_REPO>\fidle_environment_windows10.txt
conda create --name fidle --file <PATH_TO_CLONED_FIDLE_REPO>\fidle_environment_windows10_gpu_cuda101.txt

If you get some conflitcts erros, you can try the manual installation below.

Conda not found...

If the conda command is not found, the easiest way is to install Anaconda or Miniconda.

About conda environment

  • See there
  • Some infrastructures, such as IDRIS or GRICAD, offer ready-made environments. Read the docs ;-)

Manual installation

This solution can be adapted if the procedure described above does not work.
This can be the case, for example under MacOS.
List of required packages must include :

With conda you can use :

For Linux

conda create --name fidle 
conda activate fidle
conda install python=3.8 numpy=1.19 scikit-learn scikit-image matplotlib pandas jupyterlab
conda install tensorflow=2.2 
conda install -c pytorch pytorch-cpu torchvision
conda create --name fidle 
conda activate fidle
conda install python=3.8 numpy=1.19 scikit-learn scikit-image matplotlib pandas jupyterlab
conda install tensorflow-gpu=2.2 
conda install -c pytorch pytorch=1.7.1=py3.8_cuda10.1.243_cudnn7.6.3_0  torchvision

For Windows

=> with miniconda 3.8 installed and using a Anaconda prompt terminal

conda create --name fidle 
conda activate fidle
conda install python=3.8 numpy=1.19.5 scikit-learn scikit-image matplotlib pandas jupyterlab
conda install tensorflow=2.3 
conda install -c pytorch pytorch torchvision
conda create --name fidle 
conda activate fidle
conda install python=3.8 numpy=1.19.5 scikit-learn scikit-image matplotlib pandas jupyterlab
conda install tensorflow-gpu=2.3 tensorflow=2.3-mkl_py38h1fcfbd6_0 
conda install -c pytorch pytorch torchvision

This manual procedure is indicative because it can vary according to the evolution of the versions of this or that package...

Test tensorflow installation

If you want to check your tensorflow installation (and check if it supports gpu), launch a python interpreter and use these instructions :

import tensorflow as tf
print(tf.__version__)
print(tf.test.is_built_with_cuda())
print(tf.config.list_physical_devices('GPU'))

Test pytorch installation

If you want to check your pytorch installation (and check if it supports gpu), launch a python interpreter and use these instructions :

import torch
print(torch.__version__)
print(torch.cuda.is_available())
print(torch.cuda.device_count())

4/ Start Jupyter lab :

Very simply, as follows.

Remark : for Windows, use a Anaconda Prompt terminal to type these commands.

  1. Do not forget to set the FIDLE_DATASET_DIR variable

  2. Activation of the fidle environment :

conda activate fidle
  1. Start Jupyter lab :
cd <PATH_TO_CLONED_FIDLE_REPO>
jupyter lab

Comments

    Please register or sign in to add a comment.