Too many with lines in DQNTrainer and FusedGame

[honglu2875]

To time each component (very important in tweaking parameters for large scale computing), the best way I am aware so far is the with context manager like in the code. But when there are too many to time, it becomes ugly. It really hurts the readability. Is there a way to simplify this? (e.g., can one define a special syntax in comments, and convert all those comments through a single context manager outside with options to activate or not? Or is there a good profiler that suits the purpose?)

[LunaCodeDemon]

You can write a spawner class for Timer, you could give it the attributes that are commonly reused and effectively make the lines shorter.

Another thing that could be done is writing a simple wrapper for the timer stuff.
Would mean tho that you then need to put everything in lamda functions.
A wrapper for that would look something like that.

def time_that(name: str, log_dict: dict, active=True, use_cuda=False, callback, *args, **kwargs):
    with Timer(name, log_dict, active, use_cuda):
          return callback(*args, **kwargs);

So the answer is that you can remove a lot of with statements but it would make it more complicated.

[honglu2875]

Awesome! Thanks for the tips. Need to think about it.

Is it generally a bad idea to modify codes when using context manager? Ideally I would like to mark the lines I want to time in comments and only time it during profiling/debugging. In actual deployment/experiment I don't really care about those timer records.

[LunaCodeDemon]

So far i know you can remove it.
Right now i do not really know tho how to remove it via variable.
Using the spawner method you could make passthrough objects if an attribute is set, but i don't know if it's recommendable.