Python numpy array transformation

Question

I have two arrays in python, one consists of 1's the other of float values, i concatenated them and got following array:

[[0.122700 0.242400 0.000200 0.247300 0.758100 0.212200 1.000000]

 [0.276400 0.359800 0.005500 0.525200 0.787100 0.272700 1.000000]]

I would like for it to look like so:

[[0.122700, 0.242400, 0.000200, 0.247300, 0.758100, 0.212200], [1.000000]],

[[0.276400, 0.359800, 0.005500, 0.525200, 0.787100, 0.272700], [1.000000]]

since that's how the input set looks like in a NN that i'm looking to use, fragment of code of NN:

def demo():
# Teach network XOR function
testinput = [
    [[0,0], [0]],
    [[0,1], [1]],
    [[1,0], [1]],
    [[1,1], [0]]
]`

How can i accomplish this? Here's the code for the arrays:

array12 = np.ones((17,1))
array1t = np.hstack((array11color,array11diameter,array11compactness, array11asymmetry,array11solidity,array11extent))
np.set_printoptions(formatter={'float_kind':'{:f}'.format})
array2t = np.concatenate((array1t, array12), axis=1)
print(array2t)

what are the shapes of all the arrays that are referenced in your code but not provided? — Max Power
– Max Power, Commented May 1, 2017 at 15:57
array2t.shape results in (17, 7), I only put 2 rows here to make it cleaner but the columns are all there, array12.shape results in (17,1) and array1t.shape is (17,6) — Dantuzzo
– Dantuzzo, Commented May 1, 2017 at 16:06
To have a mix in element sizes, where 6 and 1, you have to either use a structured array or an object dtype array. You can't have a float array with a mix of subarray lengths. On making an object see the current question: stackoverflow.com/questions/43722023/… — hpaulj
– hpaulj, Commented May 1, 2017 at 16:51

plasmon360 · Accepted Answer · 2017-05-01 16:59:23Z

If you have array of floats and array of ones and you want to add them like the way you want, you could create new axis to array of ones and zip them together and make a new array like shown below

a = np.random.rand(2,6) # floats arrays
print a

'''
[[ 0.8481245   0.79061519  0.12447346  0.82435141  0.27577374  0.94882922]
 [ 0.1927391   0.0198176   0.2061685   0.33492186  0.29266001  0.80629359]]
'''

b = np.ones(2) # one arrays
print b

'''
[ 1.  1.]
'''

print b.shape
'''
(2,)
'''

b= b[:,None] # create a new axis for b
print b.shape # check the shape of the new b
'''
(2, 1)
'''

c = np.array(zip(a,b)) # zip a and b and make a new array
print c

'''
[[ array([ 0.8481245 ,  0.79061519,  0.12447346,  0.82435141,  0.27577374,
        0.94882922])
  array([ 1.])]
 [ array([ 0.1927391 ,  0.0198176 ,  0.2061685 ,  0.33492186,  0.29266001,
        0.80629359])
  array([ 1.])]]
'''

worked like a charm, you're my hero, i slightly modified it though and the final result did the trick: x = [[i, j] for i, j in zip(array1t,array12)]
This c is an object dtype array, with shape (2,2). Each element is an array.

Back2Basics · Accepted Answer · 2017-05-01 17:15:53Z

The question is going to be what do you want to do with it? Yes you can have that type of structure but they will be Python lists inside numpy (dtype='O'). You will lose all of the speed you are looking for in numpy operations because it will need to go out to the interpreter and evaluate each thing before acting on it. Here is how this numpy array looks like to numpy.

[[0.122700 0.242400 0.000200 0.247300 0.758100 0.212200],
 [1.000000]
 [0.276400 0.359800 0.005500 0.525200 0.787100 0.272700],
 [1.000000]]

What would an operation like transpose look like? or operations that involve diagonals?

What is typically done is these arrays of data are separated out like this:

X = np.array([[0.122700 0.242400 0.000200 0.247300 0.758100 0.212200],[0.276400 0.359800 0.005500 0.525200 0.787100 0.272700]])

Y = np.ones([2,1], dtype=np.float64)

Then you can do all the standard matrix multiplication. My concern is, when you get what you are asking for, you won't want it.

Here is a good example to follow for teaching a general linear model about XOR (at the bottom of the page)

Yeah, it's not likely that the neural network actually takes object arrays full of lists.

Collectives™ on Stack Overflow

Python numpy array transformation

2 Answers 2

2 Comments

1 Comment

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

1 Comment

Linked

Related