Skip to content

Latest commit

 

History

History
39 lines (33 loc) · 1.47 KB

anaconda_environment.md

File metadata and controls

39 lines (33 loc) · 1.47 KB
Raw

Environment

We test the project on Ubuntu 16.04(Linux), Anaconda3 (v5.2 with python3.6).

1. Setting Anaconda environment

After installing Anaconda, it is nice to create a conda environment so you do not destroy your main installation in case you make a mistake somewhere:

conda create --name tg python=3.6

Now you can switch to the new environment in your terminal by running the following (on Linux terminal):

source activate tg

2. Required Packages

PyTorch installation

Refer to PyTorch. The tested versions are v1.1.0 (py3.6_cuda9.0.176_cudnn7.5.1_0) and v0.3.0 for pytorch and torchvision, respectively. PC environment is Ubuntu 16.04 (Linux), Anaconda3 (v5.2 with python3.6).

conda install pytorch=1.1.0 torchvision=0.3.0 cudatoolkit=9.0 -c pytorch

TensorFlow for tensorboard (python 3.6)

We install CPU-version of TensorFlow since we use only tensorboard.

pip install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.8.0-cp36-cp36m-linux_x86_64.whl

Python packages

This project requires several Python packages to be installed.
You can install the packages by typing:

conda install -y nb_conda numpy scipy jupyter matplotlib pillow nltk tqdm pyyaml scikit-image scikit-learn h5py
conda install -y -c conda-forge coloredlogs
pip install moviepy