Turns Python functions into CLI via Python annotations
Project description
GearUp
Have you ever had a moment, when the code is ready, you are eager to launch it,
you want to know if your new and shiny method works or not, just to realize you need to write
100+ lines of argparse
or click
?
Gear up and get ready to go!
Quick (and only) intro
Assume your project contains main.py
script with the following functions:
def train(method : str, dataset : str, alpha : float):
<do stuff>
def test(method : str, dataset : str):
<do testing>
Just add:
from gearup import gearup
if __name__ == '__main__':
gearup(train, test)()
and you are ready to go!
> python main.py train method=resnet dataset=mnist alpha=0.01
> python main.py test method=resnet dataset=mnist
Installation
As usual:
pip install gearup
or
pip install git+https://gitlab.com/craynn/gearup.git
How it works
gearup
, applied to a function, reads signature of the function
and infers types of its arguments from the annotations:
def f(x: int, y: int):
return x + y
Annotations here can be any callable of type str -> A
,
that raises either ValueError
or TypeError
when its argument is not a proper
representation of any instance of type A
.
When gear-uped function is called without arguments it reads sys.argv
,
alternatively, it can be called with a list of strings:
gearup(f)(['1', '2']) ### result = 3
gearup(f)() ### read from console arguments
Then, gear-uped function parses arguments using the following rules:
- if
=
symbol is present in the argument:k=v
, valuev
is assigned to the argumentk
and added tokwargs
; - otherwise, the argument is treated as a positional one and appended to
args
.
After that the underlying function is called: f(*args, **kwargs)
,
converting arguments in their respective types beforehand...
Yes, no flags, no aliases, just launch script like a python function (with Haskell style)...
> python main.py 1 y=2
As a bonus, gearup.apply(f, *args, **kwargs)
provides a Python-friendly way to do the same thing, which
is useful when your script contains multiple methods with non-identical sets of parameters.
import gearup
def method1(x: int, y: int): return x + y
def method2(x: int, z: float): return x / z
def main(method: str, x: int, **kwargs):
if method == 'method1':
return gearup.apply(method1, x, **kwargs)
else:
return gearup.apply(method2, x, **kwargs)
if __name__ == '__main__':
gearup.gearup(main)()
Commands
Sometimes you need to pack several functions into one script:
gearup(train, test)()
### or
gearup(train=train, test=test)()
### or
gearup(train, test=test)()
> python main.py train <arguments for train>
> python main.py test <arguments for test>
More precisely, if supplied with more than one argument or at least one keyword argument,
gearup
consumes the first CLI argument and
switches between provided functions.
Bonus: it is recursive!
def train(...): pass
def test_fast(...): pass
def test_slow(...): pass
gearup(
train,
test=dict(
fast=test_fast,
slow=test_slow
)
)()
> python main.py train method=resnet alpha=0.1
> python main.py test slow method=resnet
Note: when a non-keyword argument is passed to gearup
,
it reads __name__
attribute of this argument. For example, gearup(f1, f2)
is equivalent to
gearup(f1=f1, f2=f2)
.
Misc.
Flags
As bool
type behaves strangely in Python (e.g., bool('False') == True
),
annotation bool
is automatically replaced by gearup.common.boolean
,
that parses strings that represent boolean values properly.
Help
Just add --help
:
> python examples/main.py --help
Available commands:
train -> (method: {nonlogreg, logreg}, power: [-2, 5), alpha: float) Trains method with alpha.
test -> slow -> (method: {nonlogreg, logreg}) Tests method...
fast -> (method: {nonlogreg, logreg, inception}) Undocumented test function.
--help
also works with commands:
> python examples/main.py test --help
Available commands:
slow -> (method: {logreg, nonlogreg}) Tests method...
fast -> (method: {logreg, inception, nonlogreg}) Undocumented test function.
> python examples/main.py test slow --help
Tests method...
A long
several lines
long
description.
(method: {nonlogreg, logreg})
Non-standard types
gearup
also defines several non-standard types:
choice(x_1, x_2, ..., x_n)
--- only accepts arguments from the provided set;either[type_1, type_2, ..., type_n]
--- tries to convert supplied value to one of the provided types; note, thattype_i
has priority overtype_j
ifi < j
, thus, e.g.,either[float, int]
is equivalent tofloat
as any string representingint
is also a validfloat
.interval[a:b]
--- half-open intervala <= x < b
, type (int or float) is inferred from types ofa
andb
; also a more complete constructor exists:interval(start, stop, left=True, right=False, cast=None)
.a < number
,a <= number
,number < b
,number <= b
- an alternative syntax for constructing intervals, intervals can also be combined via&
, e.g.,(a < number) & (number < b)
(note, that parenthesis are required as almost every operator has higher priority than comparison operators). Unfortunately, Python does not support overloading chained comparisons, thus, a nicea < number < b
syntax is not available, however,(a < number) < b
works fine.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.