This repository contains a snapshot the Jupyter notebooks I used for my 2020 master's thesis Contrastive Predictive Coding on Temporal Graphs.
The BN16 data is available at https://zenodo.org/record/3862966
Note that figures and outputs in the notebooks may not represent the findings recorded. All thesis results can, however, be reproduced.
The CPC paper reproduction code can be found in cpc_reproduction
, the take on BN16 graphs in bn16
. See the respective Readme for further details.
Dependencies for the CPC reproduction:
- python 3.7
- pytorch 1.5
- torchaudio 0.5
- cudatoolkit 10.1
- numpy 1.19.1
- pandas 1.0.5
- matplotlib 3.3.0
- jupyterlab
- torchsummary 1.5.1
Dependencies for the application on BN16:
- python 3.7
- jupyterlab
- pytorch 1.7
- numpy 1.19.4
- pandas 1.1.4
- matplotlib 3.3.2
- sparse 0.11.2
- ray[tune] 1.0.1post