CODEWITHSUNDEEP
python
Alec
Alec

Reputation: 9536

Can you pass arguments to a function object in Python?

For the sake of organization, I outline an ML optimizer with the rest of my config constants at the top of my file:

optimizer = torch.optim.SGD()

To use the optimizer, I have to pass in the model parameters, generated later on in the code

optimizer = torch.optim.SGD(model.parameters, lr=LEARNING_RATE)

Is there any way for me to pass arguments into the variable optimizer?

edit: I think my question is unclear, here's a simpler example of what I was asking:

#take the square of some arbitrary number

fn = math.prod()
x = 5

#how to feed x into the variable fn?

Upvotes: 1

Views: 48

Answers (4)

Alec
Alec

Reputation: 9536

I figured it out, seems really simple but I'll leave it up because I couldn't find it online anywhere.

You can just pass the arguments to the new variable because it's now the same object as the function

optimizer(model.parameters, lr=LEARNING_RATE)

Upvotes: 0

Techniquab
Techniquab

Reputation: 903

I'm not sure if this is exactly what you are after, but I often use functools.partial to bind values to parameters.

import functools

optimizer = functools.partial(torch.optim.SGD, model.parameters, lr=LEARNING_RATE)

then you can call optimizer as a function and it will have the first positional paramerte and the "lr" keyword parameter already set, and you can pass whatever other parameters you need when you call optimizer.

Upvotes: 0

jonesy
jonesy

Reputation: 3542

Sure! 

>>> def add(x, y):
...    return x+y
... 
>>> z = add
>>> z(1,1)
2

Upvotes: 2

chepner
chepner

Reputation: 530920

It sounds like you are asking about partial application. This is a way of "wrapping" a function and some arguments into a single, new callable, which takes fewer (or even no) arguments.

>>> from functools import partial
>>> f = lambda x, y, z: x + y + z
>>> g = partial(f, 2, 3)
>>> g(4)
9
>>> h = partial(f, 1, 2, 3)
>>> h()
6

Upvotes: 0

Related Questions