pyqol 0.0.4.6

A Pack of useful python QoL changes

PyQoL

This package contains a lot of feature for Quality of Life, functionnal programming, and others.

.CORE

codedit

Codedit is a decorator that lets you change the source code of a function using regexes. Sadly the syntax has to be "python-valid" before, but the code doesn't have to mean anything though.

@codedit("oooooooone", "1")
def add_one(x):
    return x + oooooooone
@codedit(r"_(.+)_more_than\[(.+)\]", r"\2 + \1")
def add_one(x):
    return _1_more_than[x]

Obviously, in that case it's not that useful, but I'm sure you can find some hacks using it :)

Codedits

Is a class that contains all known useful codedits.

.Lambda

Lets you define a custom lambda operator.

@Codedits.Lambda(">>")
def add_one(x):
    ld = {a >> a + 1}
    return ld(x)

Valid lambdas include: -1>, >, :, >>, ..

.Bittors (better iterators)

I is the new range

I, and its brothers IR, IC, IE are iteration functions, used to create loops. It will iterate over any iterable, and transform a int argument into a range from 0 to this int. Negative ints lead to a backwards loop. It will iterate over all arguments at the same time, and zip them.

Arguments are:

I(*args, revserse = False, enum = False, chunking=False, chunk_size=1)

Enum = True is the same as zipping with I(None), which returns an infinite loop. It also can be called with IE

Chunking, also called by IC returns multiple values at once, in a tuple.

IC(-8, chunk_size=2) -> (7, 6), (5, 4), (3, 2), (1, 0)

No start, end, step here, all arguments are the iterations

Multiterator

This lets you handle complex iteration patterns while not suffering from the usual problems caused by having nested loops. Here, an example speaks for itself:

my_iterator = Multiterator()
for i in my_iterator("outer range loop", 5):
    for j, k in my_iterator("inner zip loop", x, z):
        if ...
          my_iterator.stop("inner zip loop")
        if ...
          my_iterator.stop("outer range loop")
        if ...
          my_iterator.stop(Multiterator.all)

Multiterator's call works exactly like a I call.

.Structs

Struct

Struct takes any arguments when created, and stores them. It is similar to a JS object.

my_object = Struct(health=100, strength=20)
my_object.sword = Swords.Diamond
def _run(self: Struct):
    pass
my_object.run = _run
my_object.run()

Registry

You can create registers of functions (for plugin management, or special scoping), by creating a registry:

r = Registry()

Then, you can register functions in it, and access them that way.

@r.register
def my_happy_little_function(x):
    print(f"happy little {x}")

r.registry["my_happy_little_function"]("accident")
r["my_happy_little_function"]("programmer")

.FP

Function

You can decorate one of your functions with Function to access function composition, and other features.

@Function
def add_two(n): return n + 2
mult_by_two = Function(lambda x: x * 2)

add_then_mult = (add_two + mult_by_two)
mult_then_add = (add_two * mult_by_two)

Bunction (better function)

This class is a superset of Function, which allows for cool setups. Let's implement the fibonacci function with it, in a very defensive manner:

# First, setup the default case
@Bunction
def fib(n):
    return fib(n-1) + fib(n-2) 

Alone, this function doesn't work, it needs to return 1 if the input is 0 or 1. We can easily patch this by adding cases, which will overwrite the default.

#if x is 0, this case will be executed
@fib.case(lambda x : x == 0)
def _one(x): return 1

Yes, this is ugly, that's why you can also do this:

# if input is 1, return 1
fib.case(1)(1)

Now, let's do a little defensive programming, and make our function idiot-proof:

fib.case(lambda x : x < 0)(0)

We can even preprocess the inputs, to fit in one of our cases when it couldn't before

@fib.preprocess(lambda x : type(x) == str)
def _exec(x):
    if x.isnumber(): # implement your own isnumber, python doesn't have one for floats for some reason
        return float(x) # will then be converted to a float
    else:
        return 0 # will input 0 to fib

# Then, convert floats to ints
fib.preprocess(lambda x : type(x) == float)(lambda x : int(x))

Map

A new tool for iterating, the Map

Map.over

Map over is a curried function taking in an iterable, then a function, and outputs a new List of the results of the function. You can use it as a decorator, or as a normal function call

Map.over([0, 1, 2, 3])(lambda x : x * 2) == L(0, 2, 4, 6)
@Map.over([0, 1, 2, 3])
def newlist(e):
    return e * 2
newlist == L(0, 2, 4, 6)

That last functionnality might look extremely wierd, and it does, but it can be practical if used correctly. If you have a set of objects you iterate over everywhere in the code, that you might change, why not have them all in once place ?

agents = L(...)
ForAllAgents = Map.over(agents)
#...
@ForAllAgents
def training_log(agent):
    #...
    return log_info
print(training_log)

Map.using

It is the exact same as Map.over, but the argument order is swapped.

@Map.using
def mult_list_by_two(e):
    return e * 2
mult_list_by_two([0, 1, 2, 3]) == L(0, 2, 4, 6)

Helpers

.timer

A simple decorator for debugging the execution time of a function:

@Helpers.time
def takes_long():
    for i in I(None):
        if i == int(1e12): break;
takes_long() # will print:
# Finished takes_long in x secs

ML

A collection of helpers for making Neural Networks ! For now, most of the supported code is in PyTorch, so feel free to sumbit equivalents in other frameworks !

PyTorchModules

.Flatten

.Residual

.LowerBound

.GDN