How do I create an array where every entry is the same value? I know numpy.ones() and numpy.zeros() do this for 1's and 0's, but what about -1?
For example:
>>import numpy as np
>>np.zeros((3,3))
array([[ 0., 0., 0.],
[ 0., 0., 0.],
[ 0., 0., 0.]])
>>np.ones((2,5))
array([[ 1., 1., 1., 1., 1.],
[ 1., 1., 1., 1., 1.]])
>>np.negative_ones((2,5))
???
Use np.full() as follows:
np.full((2, 5), -1.)
Returns:
array([[-1., -1., -1., -1., -1.],
[-1., -1., -1., -1., -1.]])
I don't know if there's a nice one-liner without an arithmetic operation, but probably the fastest approach is to create an uninitialized array using empty and then use .fill() to set the values. For comparison:
>>> timeit m = np.zeros((3,3)); m += -1
100000 loops, best of 3: 6.9 us per loop
>>> timeit m = np.ones((3,3)); m *= -1
100000 loops, best of 3: 9.49 us per loop
>>> timeit m = np.zeros((3,3)); m.fill(-1)
100000 loops, best of 3: 2.31 us per loop
>>> timeit m = np.empty((3,3)); m[:] = -1
100000 loops, best of 3: 3.18 us per loop
>>> timeit m = np.empty((3,3)); m.fill(-1)
100000 loops, best of 3: 2.09 us per loop
but to be honest, I tend to either add to the zero matrix or multiply the ones matrix instead, as initialization is seldom a bottleneck.
100000 loops, best of 3: 9.49 us per loop Why does this take so long ?
[...]) or .fill are both great (and also win here). plus: readibility counts...
ones is basically empty + fill with some error catching, so the only place that any performance difference could show up is in small array sizes, because asymptotically they do exactly the same thing.
np.full(), as per my answer below.-1 * np.ones((2,5))
Multplying by the number you need in the matrix will do the trick.
In [5]: -1 * np.ones((2,5))
Out[5]:
array([[-1., -1., -1., -1., -1.],
[-1., -1., -1., -1., -1.]])
In [6]: 5 * np.ones((2,5))
Out[6]:
array([[ 5., 5., 5., 5., 5.],
[ 5., 5., 5., 5., 5.]])
For an array of -1s
-1 * np.ones((2,5))
Simply multiply with the constant.
How about:
[[-1]*n]*m
where n is the number of columns and m is the number of rows?
foo = np.repeat(10, 50).reshape((5,10))
Will create a 5x10 matrix of 10s.
According to me, these are the good way to create an array with specified value
arr=[value for x in range(num)]
or
[VALUE]*NUM
where num is the length of Array & value is the specified value.
