NumPy ndarray.all() vs np.all(ndarray) vs all(ndarray)

The difference between np.all(a) and a.all() is simple:

• If a is a numpy.array then np.all() will simply call a.all().
• If a is not a numpy.array the np.all() call will convert it to an numpy.array and then call a.all(). a.all() on the other hand will fail because a wasn't a numpy.array and therefore probably has no all method.

The difference between np.all and all is more complicated.

• The all function works on any iterable (including list, sets, generators, ...). np.all works only for numpy.arrays (including everything that can be converted to a numpy array, i.e. lists and tuples).
• np.all processes an array with specified data type, that makes it pretty efficient when comparing for != 0. all however needs to evaluate bool for each item, that's much slower.
• processing arrays with python functions is pretty slow because each item in the array needs to be converted to a python object. np.all doesn't need to do that conversion.

Note that the timings depend also on the type of your a. If you process a python list all can be faster for relativly short lists. If you process an array, np.all and a.all() will be faster in almost all cases (except maybe for object arrays, but I won't go down that path, that way lies madness).

I'll take a swing at this

• np.all is a generic function which will work with different data types, under the hood this probably looks for ndarray.all which is why it's slightly slower.

• all is a python bulit-in function see https://docs.python.org/2/library/functions.html#all.

• ndarray.all is method of the 'numpy.ndarray' object, calling this directly may be faster.

I suspect that numpy's functions do more to evaluate an array element as a boolean, likely in some generic numeric-first way, while the builtin all() does nothing, since the elements are already booleans.

I wonder how different the results would be with integers of floats.