`np.mean()` on list to `sum(ls) / len(ls)`

[baziotis]

Original

def foo(x):
  return np.mean([len(w) for w in str(x).split()])
df["A"] = df["B"].apply(foo)

Rewritten

def foo(x):
  ls = [len(w) for w in str(x).split()]
  return sum(ls) / len(ls)
df["A"] = df["B"].apply(foo)

This can be difficult to generalize. Let's just say for now that if we have np.mean(<list comprehension>), we transform it as above.

Note: Usually, numpy functions should be applied only to numpy arrays, otherwise they are slower than the Python ones.