In python, what is the difference between 'import foo.bar as bar' and 'from foo import bar'?

Question

I see this convention in pyTorch and matplotlib:

import torch.nn as nn
import torch.optim as optim

import matplotlib.pyplot as plt

Is there a reason why the whole path (module.submodule) is being imported as an alias instead of just the submodule? What is the difference if I import like this:

from torch import nn
from torch import optim

from matplotlib import pyplot as plt

Edit: So for the generic case:

import foo.bar as bar    # [1]
from foo import bar      # [2]

Can there exist code which refers to bar such that it will run with [1] and not [2] (and vice versa)? I.e. is there functional difference between these two ways to import?

Answer 1

Behind the scenes, all of the import statements are essentially mapped to the builtin __import__ eg:

import torch.nn as nn

becomes

nn = __import__("torch.nn", globals(), locals(), [], 0)

similarly:

from torch import nn

becomes

nn = __import__("torch", globals(), locals(), ["nn"], 0).nn

As for matplotlib:

import matplotlib.pyplot as plt
# becomes
plt = __import__("matplotlib.pyplot", globals(), locals(), [], 0)

from matplotlib import pyplot as plt
# becomes
plt = __import__("matplotlib", globals(), locals(), ['pyplot'], 0).pyplot

Subtly different but functionally equivalent.

Reference: https://docs.python.org/3/library/functions.html#__import__