Kproject

kproject is initializing tool to clean repo and to wirte experiment code quickly.

Tutorial

First, you use new command of kproject and create new some directories and files such.

$ kproject new mnist
$ cd mnist; ls -R
README.md     dataset       kproject.json model         result        src

./dataset:

./model:

./result:

./src:
lib     main.py

./src/lib:
args.py

Here is recommend usage of these directories.

name	recommend way to use
dataset	Save some dataset you use.
models	Save your models here.
results	Save your experiment results.
src	Save your codes.
src/lib	Save your codes of library.

Here is description some files created automatically.

name	description
README.md	You should know.
kproject.json	Automatically use with `run` command
src/main.py	You should write main codes here.
src/lib/args.py	This file says code of argument parser.

Next, you use run command.

$ kproject run
[ INFO ] Run in `kproject.json`

Automatically run src/main.py with configuration of kproject.json. Now, you implicitly used kprojec.json, so let's see more in detail. Here is content of initial krpoject.json.

{
    "experiments": [
        {
            "cmd": "./src/main.py",
            "exec?": true,
	    "args": {
	    
	    }
	}
    ]
}

If you added command line argument and use the argument, you should set value of args.

Useful kproject library

kproject has some useful library. Here is useful library I think.

About argument parser

Already template argument parser has implemented by argparse in src/lib/args.py. So you can easy to add some options. Then this file defined CONFIG.

from lib.args import CONFIG
print(CONFIG['title'])

About logging

Some logging class has already defined.

from kproject.log import MDLog
from kproject.utils import *
log = MDLog("experiment1") #=> create /result/experiment1.md
log.start()
log.info("some information")
log.info(h2("experiment 1-1")) #=> ## experiment 1-1
log.finish()

Install

$ pip install git+https://github.com/owlinux1000/kproject