Given a t1xt2xn array and a m1xm2 mask, how to obtain the t1xt2xn array where the n-dim arrays are convolved with the mask?
The function scipy.signal.convolve is not able to handle this because it only accept inputs with same number of dimensions.
Example with the "same" logic:
in1 =
[[[0,1,2],[3,4,5],[6,7,8]],
[[9,10,11],[12,13,14],[15,16,17]],
[[18,19,20],[21,22,23],[24,25,26]]]
in2 =
[[0,1,0],
[0,1,0],
[0,1,0]]
output =
[[[0,0,0],[15,17,19],[0,0,0]],
[[0,0,0],[36,39,42],[0,0,0]],
[[0,0,0],[33,35,37],[0,0,0]]]
I'm very sorry but I haven't strong math background, so my answer could be wrong. Anyway, if you need to use mask for selecting you should convert it to bool type. For example:
in1 = np.array([[[0,1,2], [3,4,5], [6,7,8]],
[[9,10,11], [12,13,14], [15,16,17]],
[[18,19,20], [21,22,23], [24,25,26]]])
in2 = np.array([[0, 1, 0],
[0, 1, 0],
[0, 1, 0]])
mask = in2.astype(bool)
print(in1[mask])
# [[ 3 4 5]
# [12 13 14]
# [21 22 23]]
in3 = np.zeros(in1.shape)
in3[mask] = np.convolve(in1[mask].ravel(), in2.ravel(), 'same').reshape(mask.shape)
print(in3)
# [[[ 0. 0. 0.]
# [ 15. 17. 19.]
# [ 0. 0. 0.]]
#
# [[ 0. 0. 0.]
# [ 36. 39. 42.]
# [ 0. 0. 0.]]
#
# [[ 0. 0. 0.]
# [ 33. 35. 37.]
# [ 0. 0. 0.]]]
I'm not very sure about last part, especially about reshaping but I hope you get an idea.
Related
I'm trying to create a matrix that reads:
[0,1,2]
[3,4,5]
[6,7,8]
However, my elements keep repeating. How do I fix this?
import numpy as np
n = 3
X = np.empty(shape=[0, n])
for i in range(3):
for j in range(1,4):
for k in range(1,7):
X = np.append(X, [[(3*i) , ((3*j)-2), ((3*k)-1)]], axis=0)
print(X)
Results:
[[ 0. 1. 2.]
[ 0. 1. 5.]
[ 0. 1. 8.]
[ 0. 1. 11.]
[ 0. 1. 14.]
[ 0. 1. 17.]
[ 0. 4. 2.]
[ 0. 4. 5.]
I'm not really sure how you think your code was supposed to work. You are appending a row in X at each loop, so 3 * 3 * 7 times, so you end up with a matrix of 54 x 3.
I think maybe you meant to do:
for i in range(3):
X = np.append(X, [[3*i , 3*i+1, 3*i+2]], axis=0)
Just so you know, appending array is usually discouraged (just create a list of list, then make it a numpy array).
You could also do
>> np.arange(9).reshape((3,3))
array([[0, 1, 2],
[3, 4, 5],
[6, 7, 8]])
columns = np.shape(lines)[0] # Gets x-axis dimension of array lines (to get numbers of columns)
lengths = np.zeros(shape=(2,1)) # Create a 2D array
# lengths = [[ 0.]
# [ 0.]]
lengths = np.arange(columns).reshape((columns)) # Makes array have the same number of columns as columns and fills it with elements going up from zero <--- This line seems to be turning it into a 1D array
Output after printing lengths array:
print(lengths)
[0 1 2]
Expected Output Example:
print(lengths)
[[0 1 2]] # Notice the double square bracket
This results in me not being able to enter data into a 2D parts of an array, because it now no longer exists:
np.append(lengths, 65, axis=1)
AxisError: axis 1 is out of bounds for array of dimension 1
I want the array to be 2D so I can store "IDs" on the first row and values on the second (at a later point in the program). I'm also aware that I could add another row to the array instead of doing it at initialization. But I'd rather not do that since I heard that's inefficient and this program's success is highly dependent on performance.
Thank you.
Since you eventually want a 2d array with ids in one row and values in the second, I'd suggest starting with the right size
In [535]: arr = np.zeros((2,10),int)
In [536]: arr
Out[536]:
array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0]])
In [537]: arr[0,:]=np.arange(10)
In [538]: arr
Out[538]:
array([[0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0]])
Sure you could start with a 1 row array of ids, but adding that 2nd row at a later time requires making a new array anyways. np.append is just a variation on np.concatenate.
But to make a 2d array from arange I like:
In [539]: np.arange(10)[None,:]
Out[539]: array([[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]])
reshape also works, but has to be given the correct shape, e.g. (1,10).
In:
lengths = np.zeros(shape=(2,1)) # Create a 2D array
lengths = np.arange(columns).reshape((columns))
the 2nd lengths assignment replaces the first. You have to do an indexed assignment as I did with arr[0,:] to modify an existing array. lengths[0,:] = np.arange(10) wouldn't work because lengths only has 1 column, not 10. Assignments like this require correct pairing of dimensions.
Don't need 2D data to put into a column of a 2D array. You just need 1D data.
You can put the data into the 0th row instead of the 0th column if you change the organization of memory. This is copying data into contiguous memory (memory without gaps) and that is faster.
Program:
import numpy as np
data = np.arange(12)
#method 1
buf = np.zeros((12, 6))
buf[:,0] = data
print(buf)
#method 2
buf = np.zeros((6, 12))
buf[0] = data
print(buf)
Result:
[[ 0. 0. 0. 0. 0. 0.]
[ 1. 0. 0. 0. 0. 0.]
[ 2. 0. 0. 0. 0. 0.]
[ 3. 0. 0. 0. 0. 0.]
[ 4. 0. 0. 0. 0. 0.]
[ 5. 0. 0. 0. 0. 0.]
[ 6. 0. 0. 0. 0. 0.]
[ 7. 0. 0. 0. 0. 0.]
[ 8. 0. 0. 0. 0. 0.]
[ 9. 0. 0. 0. 0. 0.]
[ 10. 0. 0. 0. 0. 0.]
[ 11. 0. 0. 0. 0. 0.]]
[[ 0. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]]
How does the np.newaxis work within the index of numpy array in program 1? Why it works like this?
Program 1:
import numpy as np
x_id = np.array([0, 3])[:, np.newaxis]
y_id = np.array([1, 3, 4, 7])
A = np.zeros((6,8))
A[x_id, y_id] += 1
print(A)
Result 1:
[[ 0. 1. 0. 1. 1. 0. 0. 1.]
[ 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 1. 0. 1. 1. 0. 0. 1.]
[ 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0.]]
The newaxis turns the x_id array into a column vector, same as np.array([[0],[3]]).
So you are indexing A at the cartesian product of [0,3] and [1,3,4,7]. Or to put it another way, you end up with 1s for rows 0 and 3, columns 1,3,4,and 7.
Also look at np.ix_([0,3], [1,3,4,7])
or
In [832]: np.stack(np.meshgrid([0,3],[1,3,4,7],indexing='ij'),axis=2)
Out[832]:
array([[[0, 1],
[0, 3],
[0, 4],
[0, 7]],
[[3, 1],
[3, 3],
[3, 4],
[3, 7]]])
Be a little careful with the +=1 setting; if indices are duplicated you might not get what you expect, or would get with a loop.
I have a numpy array of shape [12, 8, 5, 5]. I want to modify the values of 3rd and 4th dimension for each element.
For e.g.
import numpy as np
x = np.zeros((12, 80, 5, 5))
print(x[0,0,:,:])
Output:
[[ 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0.]]
Modify values:
y = np.ones((5,5))
x[0,0,:,:] = y
print(x[0,0,:,:])
Output:
[[ 1. 1. 1. 1. 1.]
[ 1. 1. 1. 1. 1.]
[ 1. 1. 1. 1. 1.]
[ 1. 1. 1. 1. 1.]
[ 1. 1. 1. 1. 1.]]
I can modify for all x[i,j,:,:] using two for loops. But, I was wondering if there is any pythonic way to do it without running two loops. Just curious to know :)
UPDATE
Actual use case:
dict_weights = copy.deepcopy(combined_weights)
for i in range(0, len(combined_weights[each_layer][:, 0, 0, 0])):
for j in range(0, len(combined_weights[each_layer][0, :, 0, 0])):
# Extract 5x5
trans_weight = combined_weights[each_layer][i,j]
trans_weight = np.fliplr(np.flipud(trans_weight ))
# Update
dict_weights[each_layer][i, j] = trans_weight
NOTE: The dimensions i, j of combined_weights can vary. There are around 200 elements in this list with varied i and j dimensions, but 3rd and 4th dimensions are always same (i.e. 5x5).
I just want to know if I can updated the elements combined_weights[:,:,5, 5] with transposed values without running 2 for loops.
Thanks.
Simply do -
dict_weights[each_layer] = combined_weights[each_layer][...,::-1,::-1]
I have a numpy array D of dimensions 4x4
I want a new numpy array based on an user defined value v
If v=2, the new numpy array should be [D D].
If v=3, the new numpy array should be [D D D]
How do i initialise such a numpy array as numpy.zeros(v) dont allow me to place arrays as elements?
If I understand correctly, you want to take a 2D array and tile it v times in the first dimension? You can use np.repeat:
# a 2D array
D = np.arange(4).reshape(2, 2)
print D
# [[0 1]
# [2 3]]
# tile it 3 times in the first dimension
x = np.repeat(D[None, :], 3, axis=0)
print x.shape
# (3, 2, 2)
print x
# [[[0 1]
# [2 3]]
# [[0 1]
# [2 3]]
# [[0 1]
# [2 3]]]
If you wanted the output to be kept two-dimensional, i.e. (6, 2), you could omit the [None, :] indexing (see this page for more info on numpy's broadcasting rules).
print np.repeat(D, 3, axis=0)
# [[0 1]
# [0 1]
# [0 1]
# [2 3]
# [2 3]
# [2 3]]
Another alternative is np.tile, which behaves slightly differently in that it will always tile over the last dimension:
print np.tile(D, 3)
# [[0, 1, 0, 1, 0, 1],
# [2, 3, 2, 3, 2, 3]])
You can do that as follows:
import numpy as np
v = 3
x = np.array([np.zeros((4,4)) for _ in range(v)])
>>> print x
[[[ 0. 0. 0. 0.]
[ 0. 0. 0. 0.]
[ 0. 0. 0. 0.]
[ 0. 0. 0. 0.]]
[[ 0. 0. 0. 0.]
[ 0. 0. 0. 0.]
[ 0. 0. 0. 0.]
[ 0. 0. 0. 0.]]
[[ 0. 0. 0. 0.]
[ 0. 0. 0. 0.]
[ 0. 0. 0. 0.]
[ 0. 0. 0. 0.]]]
Here you go, see if this works for you.
import numpy as np
v = raw_input('Enter: ')
To intialize the numpy array of arrays from user input (obviously can be whatever shape you're wanting here):
b = np.zeros(shape=(int(v),int(v)))
I know this isn't initializing a numpy array but since you mentioned wanting an array of [D D] if v was 2 for example, just thought I'd throw this in as another option as well.
new_array = []
for x in range(0, int(v)):
new_array.append(D)