Skip to main content

REProducible Experimental environment.

Project description

fret

Travis (.org) Coveralls github PyPI PyPI - Python Version

Framework for Reproducible ExperimenTs

Installation

From pip:

pip install fret

From source: clone the repository and then run: python setup.py install.

Tutorial

Basic Usage

Create a file named app.py with content:

import fret

@fret.command
def run(ws):
    model = ws.build()
    print(model)

@fret.configurable
class Model:
    def __init__(self, x=3, y=4):
        ...

Then under the same directory, you can run:

$ fret config Model
[ws/_default] configured "main" as "Model" with: x=3, y=4
$ fret run
Model(x=3, y=4)
$ fret config Model -x 5 -y 10
[ws/_default] configured "main" as "Model" with: x=5, y=10
$ fret run
Model(x=5, y=10)

Using Workspace

You can specify different configuration in different workspace:

$ fret -w ws/model1 config Model
[ws/model1] configured "main" as "Model" with: x=3, y=4
$ fret -w ws/model2 config Model -x 5 -y 10
[ws/model2] configured "main" as "Model" with: x=5, y=10
$ fret -w ws/model1 run
Model(x=3, y=4)
$ fret -w ws/model2 run
Model(x=5, y=10)

You can ommit -w <path> if you are currently under a workspace:

$ cd ws/model2
$ fret run
Model(x=5, y=10)

Save/Load

import fret

@fret.command
def train(ws):
    model = ws.build()
    model.train()
    ws.save(model, 'trained')

@fret.command
def test(ws):
    model = ws.load('ws/best/snapshot/main.trained.pt')
    print(model.weight)

@fret.configurable
@fret.stateful('weight')
class Model:
    def __init__(self):
        self.weight = 0
    def train(self):
        self.weight = 23
$ fret -w ws/best config Model
[ws/_default] configured "main" as "Model"
$ fret -w ws/best train
$ fret test
23

An Advanced Workflow

In app.py:

import time
import fret

@fret.configurable
@fret.stateful('value')
class Model:
    def __init__(self):
        self.value = 0

@fret.command
def resumable(ws):
    model = ws.build()
    with ws.run('exp-1') as run:
        run.register(model)
        cnt = run.acc()
        for e in fret.nonbreak(run.range(5)):
            # with `nonbreak`, the program always finish this loop before exit
            model.value += e
            time.sleep(0.2)
            cnt += 1
            print('current epoch: %d, sum: %d, cnt: %d' %
                  (e, model.value, cnt))

Then you can stop and restart this program anytime, with consistent results:

$ fret resumable
current epoch: 0, sum: 0, cnt: 1
current epoch: 1, sum: 1, cnt: 2
^CW SIGINT received. Delaying KeyboardInterrupt.
current epoch: 2, sum: 3, cnt: 3
Traceback (most recent call last):
    ...
KeyboardInterrupt
W cancelled by user
$ fret resumable
current epoch: 3, sum: 6, cnt: 4
current epoch: 4, sum: 10, cnt: 5

Submodule

@fret.configurable
class A:
    def __init__(self, foo):
        ...

@fret.configurable(submodules=['sub'])
class B:
    def __init__(self, sub, bar=3):
        self.sub = sub(foo='bar')   # call sub to build submodule
$ fret config sub A
[ws/_default] configured "sub" as "A"
$ fret config B
[ws/_default] configured "main" as "B" with: sub='sub', bar=3
$ fret run
B(sub=A(), bar=3)

Inheritance

@fret.configurable
class A:
    def __init__(self, foo='bar', sth=3):
        ...

@fret.configurable
class B(A):
    def __init__(self, bar=3, **others):
        super().__init__(**others)
        ...
$ fret config B -foo baz -bar 0
[ws/_default] configured "main" as "B" with: bar=0, foo='baz', sth=3
$ fret run
B(bar=0, foo='baz', sth=3)

Internals

>>> config = fret.Configuration({'foo': 'bar'})
>>> config
foo='bar'

TODO

  • fret.Configuration: high-level class for configuration
  • fret.Workspace: module build/save/load (by tag or by path), persistency
  • ws.run() context manager, run.value(), run.acc(), run.range(), run.register()
  • @fret.configurable: parameter checking
  • fret.App, global app object
  • CLI: entry point logic, testing, tagged workspace
  • Java/GNU style command line args, shorthands, better logic for boolean default value
  • fret.toml: global configuration, configuration by class name
  • Documents and examples
  • fret new command with interactive CLI
  • Other fret commands like show log, check module, etc.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

fret-0.2.0.tar.gz (14.1 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

fret-0.2.0-py3-none-any.whl (15.8 kB view details)

Uploaded Python 3

File details

Details for the file fret-0.2.0.tar.gz.

File metadata

  • Download URL: fret-0.2.0.tar.gz
  • Upload date:
  • Size: 14.1 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/1.12.1 pkginfo/1.4.2 requests/2.19.1 setuptools/39.1.0 requests-toolbelt/0.8.0 tqdm/4.28.1 CPython/3.6.5

File hashes

Hashes for fret-0.2.0.tar.gz
Algorithm Hash digest
SHA256 66594e3b136dc9804686091e193736f4d2fd0c2e7a1893b27861f76cedd6b366
MD5 063e1fa0076020d1acedc1d2f70bfdca
BLAKE2b-256 8d66119cc41bc670f44fcdb4c1fd9b156623635ad6d14491dc252aeecab47ab4

See more details on using hashes here.

File details

Details for the file fret-0.2.0-py3-none-any.whl.

File metadata

  • Download URL: fret-0.2.0-py3-none-any.whl
  • Upload date:
  • Size: 15.8 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/1.12.1 pkginfo/1.4.2 requests/2.19.1 setuptools/39.1.0 requests-toolbelt/0.8.0 tqdm/4.28.1 CPython/3.6.5

File hashes

Hashes for fret-0.2.0-py3-none-any.whl
Algorithm Hash digest
SHA256 73d50c950fd3874c48cf44b3208640842df8c5c3f891cf72360bb68ade9f8ed3
MD5 b33fa2ae829692029c2960858cf8a78f
BLAKE2b-256 5ba847a9fbc5ef44d71fbdac04f3dd6f67848fc0fe14421327c53f507b559879

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page