https://github.com/c-hofer/topologically_densified_distributions

https://github.com/c-hofer/topologically_densified_distributions

Last synced: 4 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/c-hofer/topologically_densified_distributions
• Owner: c-hofer
• Created: 2020-03-02T10:41:54.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2020-09-30T09:12:43.000Z (over 3 years ago)
• Last Synced: 2024-01-16T10:46:22.583Z (5 months ago)
• Language: Python
• Size: 3.68 MB
• Stars: 2
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Lists

• awesome-stars - c-hofer/topologically_densified_distributions - (Python)

README

        
This repository contains the code for our work **[Topologically Densified Distributions](https://arxiv.org/abs/2002.04805)** which was presented at ICML'20.

# Installation

In the following `` will be the directory you have chosen for the installation.

1. Install Anaconda from [here](https://repo.anaconda.com/archive/Anaconda3-2020.07-Linux-x86_64.sh) into `/anaconda3`, i.e., set the prefix accordingly in the installer. Do **not** initialize your shell via the installer (the installer asks you this at the end of the installation). 

2. Activate Anaconda installation: 

    ```

    eval "$(/anaconda3/bin/conda shell.bash hook)"

    ```

3. Install pytorch via conda

    ```

    conda install torchvision cudatoolkit= -c pytorch

    ```

4. Install other dependencies

    ```

    pip install fastprogress

    ```

5. Install `torchph` via 

    ```

    cd 

    git clone -b 'submission_icml2020' --single-branch --depth 1 https://github.com/c-hofer/torchph.git

    conda develop torchph

    ```

6. Clone this repository into ``. 

# Application

1. Use the `run_experiments.py` script to run experiments. Pre-configured is an experiment on `cifar10` with the proposed regularization. Alter the script to run different experiments (see `run_experiments.py`). 

If you run the script, each experiment gets a unique id and its output is written into a sub-folder for the `results` directory. 

2. The notebook `explore_results.ipynb` contains some code to browse the results.