Add arguments to Python decorators

Reading time ~1 minute

Python decorator are a convenient way to wrap a function with another one. Per example, when timing a function, it is nice to call the timer before and after without having to rewrite the same timer.start() and timer.stop() calls everywhere.

Among other nice features of decorators, there are a couple of them which I want to stress here:

  • they apply to classes,
  • they accept arguments.

Here, I want to implement a printer for my classes showing exactly which class has called some of its methods and when. Ideally, the colors should be specified for each class, so that it makes it easy to monitor the execution of the code when various objects called different methods.

This is something I often face when dealing with machine learning pipelines: they consist of many objects (blend of models, feature pipelines) and it is hard to know exactly who is running an when, and identify bottlenecks at a glance.

Class decorators

Applying a decorator to a class allows to add method on classes directly, a little bit like inheritance. It drove my puzzled: why would one prefer a class decorator to inheritance ? See per example this question. The thing is that inheritance actually has a meaning in terms of your object: a cat is an animal, therefore, inheritance makes sense. On the other hand, if you want to add generic helpers that do not really fit into your object definitions, a class decorator is preferable.

import datetime

def class_printer(cls):

      def print_with_time(self, s):
          now_str ="%m/%d/%y, %h:%m:%s")
          print(f"[{type(self).__name__} - {now_str}] {s}")

      setattr(cls, 'print_with_time', print_with_time)

      return cls

class a:

    def __init__(self):

    def run(self):
        self.print_with_time("run method called")

if __name__ == "__main__":
    a = a()

And it outputs:

[A - 03/08/2022, 11:20:44] run method called

Great! Now, every time we need this specific printer, we don’t have much to do. Time to improve it!

Passing arguments to a decorator


import datetime

class bcolors:
    HEADER = '\033[95m'
    OKBLUE = '\033[94m'
    OKCYAN = '\033[96m'
    OKGREEN = '\033[92m'
    WARNING = '\033[93m'
    FAIL = '\033[91m'
    ENDC = '\033[0m'
    BOLD = '\033[1m'
    UNDERLINE = '\033[4m'

def color_class_printer(color):
    def class_printer(cls):

        def print_with_time(self, s):
            now_str ="%m/%d/%Y, %H:%M:%S")
            print(f"[{color}{type(self).__name__} - {now_str}{bcolors.ENDC}] {s}")

        setattr(cls, 'print_with_time', print_with_time)

        return cls
    return class_printer

OCaml List rev_map vs map

If you found this page, you are probably very familiar with OCaml already!So, OCaml has a ````map```` function whose purpose is pretty cl...… Continue reading

How to optimize PyTorch code ?

Published on March 17, 2024

Acronyms of deep learning

Published on March 10, 2024