DFRUN

gadget for dl experiments

What it does:

1. create a folder to copy all the code in the current directory to freeze one training process
2. switch to the working path to the folder created, and run the command
3. if gpu resources is needed, nvidia-ml-py package is needed. In python 2.x, pip install nvidia-ml-py, while in python 3.x, pip install nvidia-ml-py3

Dependencies:

​ Code is tested on python 3.4 and it's supposed to run in python 3 environment.

Install:

pip install dfrun

or clone this repository and run:

python setup.py install

Usage:

dfrun -d [experiment folder] -n [experiment name] -g [NUMxMEM] [command]

What does the project folder look like:

Proj

|__ [CODE FILES]

|__ .dfignore (contains the files/folders ignored in the copy process)

|__ Experiment Folder (--exp-dir -d)

​ |__ Experiment Name (--exp-name -n)

​ | |__ [CODE FILES COPIED]

​ | Other Exp Names....

Contact:

[email protected]