https://github.com/csinva/dnn-experiments
A set of scripts and experiments making it easier to analyze deep learning empirically.
https://github.com/csinva/dnn-experiments
ai artificial-intelligence cifar computer-vision deep-learning machine-learning ml mnist neural-network polynomial python pytorch sparse-coding statistics theory
Last synced: 3 months ago
JSON representation
A set of scripts and experiments making it easier to analyze deep learning empirically.
- Host: GitHub
- URL: https://github.com/csinva/dnn-experiments
- Owner: csinva
- Created: 2018-07-31T23:07:30.000Z (almost 7 years ago)
- Default Branch: master
- Last Pushed: 2020-06-13T22:39:43.000Z (about 5 years ago)
- Last Synced: 2025-02-08T13:43:05.993Z (5 months ago)
- Topics: ai, artificial-intelligence, cifar, computer-vision, deep-learning, machine-learning, ml, mnist, neural-network, polynomial, python, pytorch, sparse-coding, statistics, theory
- Language: Jupyter Notebook
- Homepage:
- Size: 42 MB
- Stars: 2
- Watchers: 5
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: readme.md
Awesome Lists containing this project
README
# understanding how deep learning works
this repo contains code for running a variety of different experiments attempting to understand deep learning via empirical experiments
# organization
- each folder contains a readme with code documentation, as well as comments in the code
- the vision_fit and vision_analyze folders detail a number of experiments on multilayer perceptrons and convolutional neural networks using various datasets including MNIST, CIFAR, and custom datasets
- the sparse_coding folder contains code for running and analyzing sparse coding on different sets of images
- the mog folder contain code examples for fitting synthetic datasets generated as mixtures of Gaussians
- the poly_fit folder contains code for fitting simple 1D polynomials
- the scripts folder contains scripts for launching jobs on a slurm cluster
- the eda folder contains minimum working examples for simple setups with various pytorch and scikit-learn functions
# requirements
- the code is all tested in python3 and pytorch 1.0
# running
- the `scripts` folder contains sample slurm scripts for launching jobs on a cluster
- most of the experiments are time-consuming and should be parallelized over many machines
- to do so, ssh into one of the scf nodes (e.g. legolas) and run ```module load python```
- set the parameters you want to sweep as lists in one of the submit*.py files
- then run this file and it will automatically launch slurm jobs for each set of parameters