https://github.com/alfa-group/reckless-minimax

[LeGO/GOW 2018] "On the Application of Danskin’s Theorem to Derivative-Free Minimax Optimization" by Abdullah Al-Dujaili, Shashank Srikant, Erik Hemberg, Una-May O'Reilly

https://github.com/alfa-group/reckless-minimax

Science Score: 10.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • CITATION.cff file
  • codemeta.json file
  • .zenodo.json file
  • DOI references
  • Academic publication links
    Links to: arxiv.org
  • Academic email domains
  • Institutional organization owner
  • JOSS paper metadata
  • Scientific vocabulary similarity
    Low similarity (11.0%) to scientific vocabulary

Keywords

black-box-optimization minimax
Last synced: 10 months ago · JSON representation

Repository

[LeGO/GOW 2018] "On the Application of Danskin’s Theorem to Derivative-Free Minimax Optimization" by Abdullah Al-Dujaili, Shashank Srikant, Erik Hemberg, Una-May O'Reilly

Basic Info
  • Host: GitHub
  • Owner: ALFA-group
  • License: mit
  • Language: Python
  • Default Branch: master
  • Homepage:
  • Size: 787 KB
Statistics
  • Stars: 1
  • Watchers: 4
  • Forks: 2
  • Open Issues: 1
  • Releases: 0
Topics
black-box-optimization minimax
Created almost 8 years ago · Last pushed almost 8 years ago

https://github.com/ALFA-group/reckless-minimax/blob/master/

# reckless-minimax
code for [On the Application of Danskins Theorem to Derivative-Free Minimax Optimization](https://arxiv.org/pdf/1805.06322.pdf)


### Installation:

- `environment.yml` lists the package dependencies. If you have `conda`:
```
conda env create -f ./environment.yml
```
and then activate the environment
```
source activate reckless
```



### Running Experiments:

`cd` to the main directory:

```
export PYTHONPATH=.
python experiments/es_experiment.py
```

This would run experiments for ES variants. Likewise, `feval_experiments.py` is for convergence experiments (regret vs. function evalutions), `budget_experiment.py` is for steps along the decent direction, and `scale_experiment.py` is for scalability experiments. Experiments results are stored under `experiments/results/` in the form of json files

To generate figures of the papers:

```
python utils/generate_plots.py

```

Figures will be generated under `experiments/results/figs/` corresponding to json files in `experiments/results`

### Statistical validity of experiments

The statstical difference between experiments from different datasets and techniques is measured using the Nemenyi test, at a signficance level of 0.05 [1]

`Orange`, a data-mining library has been used to calculate the critical difference (CD) measures and generate their plots. Specifically, the `graph_ranks` method from `Orange.evaluation.scoring` generates the CD-plots shown in our paper.

The plotting script can be found at `utils/plot_cd.py` 

#### Reference
[1] Demar, Janez. "Statistical comparisons of classifiers over multiple data sets." Journal of Machine learning research 7.Jan (2006): 1-30.

Owner

  • Name: Anyscale Learning For All (ALFA)
  • Login: ALFA-group
  • Kind: organization
  • Email: alfa-apply@csail.mit.edu
  • Location: Cambridge, MA, USA

Scalable machine learning technology, Adversarial AI, Evolutionary algorithms, and data science frameworks.

GitHub Events

Total
Last Year

Dependencies

environment.yml pypi
  • cma *