CODEWITHSUNDEEP
pythonnumpy
Suraj Kothari
Suraj Kothari

Reputation: 1632

NumPy empty() array not giving random float after defining normal NumPy array

I was using the NumPy np.empty() to get an array with a random value, but it doesn't work when I define a normal np.array() before.

Here is the two functions I used:

import numpy as np

def create_float_array(x):
    return np.array([float(x)])

def get_empty_array():
    return np.empty((), dtype=np.float).tolist()

Just to test the get_empty_array(), I wrote in the console:

>>> get_empty_array() # Should return a random float
>>> 0.007812501848093234

I was pleased with the result, so I tried this, but it didn't work the way I wanted:

>>> create_float_array(3.1415) # Create a NumPy array with the float given
>>> array([3.1415])
>>> get_empty_array() # Should return another random value in a NumPy array
>>> 3.1415

I am not too sure as to why creating a NumPy array affects the np.empty() method from giving a random value. Apparently, it gives the same value as the value in the np.array(), in this case, 3.1415.

Note that I chose to leave the shape of the np.empty() to nothing for testing purposes, but in reality it would have some shape.

Finally, I know this is not the correct way of getting random values, but I need to use the np.empty() in my program, but don't exactly know why this behaviour occurs.

Upvotes: 0

Views: 1212

Answers (1)

MSeifert
MSeifert

Reputation: 152647

Just to clarify the point:

np.empty is not giving truly random values. The offical NumPy documentation states that it will contain "uninitialized entries" or "arbitary data":

numpy.empty(shape, dtype=float, order='C')

Return a new array of given shape and type, without initializing entries.

[...]

Returns:

out : ndarray

Array of uninitialized (arbitrary) data of the given shape, dtype, and order. Object arrays will be initialized to None.

So what does uninitialized or arbitrary mean? To understand that you have to understand that when you create an object (any object) you need to ask someone (that someone can be the NumPy internals, the Python internals or your OS) for the required amount of memory.

So when you create an empty array NumPy asks for memory. The amount of memory for a NumPy array will be some overhead for the Python object and a certain amount of memory to contain the values of the array. That memory may contain anything. So an "uninitialized value" means that it simply contains whatever is in that memory you got.

What happened here is just a coincidence. You created an array containing one float, then you printed it, then it is destroyed again because you noone kept a reference to it (although that is CPython specific, other Python implementations may not free the memory immediately, they just free it eventually). Then you create an empty array containing one float. The amount of memory for the second array is identical to the amount of memory just released by the first memory. Here's where the coincidence comes in: So maybe something (NumPy, Python or your OS) decided to give you the same memory location again.

Upvotes: 4

Related Questions