There is no array type in python, but to emulate it we can use lists. I want to have 2d array-like structure filled in with zeros. My question is: what is the difference, if any, in this two expressions:
zeros = [[0 for i in xrange(M)] for j in xrange(M)]
and
zeros = [[0]*M]*N
Will zeros be same? which one is better to use by means of speed and readability?
You should use numpy.zeros. If that isn't an option, you want the first version. In the second version, if you change one value, it will be changed elsewhere in the list -- e.g.:
>>> a = [[0]*10]*10
>>> a
[[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], [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]]
>>> a[0][0] = 1
>>> a
[[1, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0, 0, 0, 0, 0], [1, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
This is because (as you read the expression from the inside out), you create a list of 10 zeros. You then create a list of 10 references to that initial list of 10 zeros.
Note that:
zeros = [ [0]*M for _ in range(N) ] # Use xrange if you're still stuck in the python2.x dark ages :).
will also work and it avoids the nested list comprehension. If numpy isn't on the table, this is the form I would use.
for Python 3 (no more xrange), the preferred answer
zeros = [ [0] * N for _ in range(M)]
for M x N array of zeros
In second case you create a list of references to the same list. If you have code like:
[lst] * N
where the lst is a reference to a list, you will have the following list:
[lst, lst, lst, lst, ..., lst]
But because the result list contains references to the same object, if you change a value in one row it will be changed in all other rows.
Zhe Hu's answer is the safer one and should have been the best answer. This is because if we use the accepted answer method
a = [[0] * 2] * 2
a[0][0] = 1
print(a)
will give the answer
[[1,0],[1,0]]
So even though you just want to update the first row first column value, all the values in the same column get updated. However
a = [[0] * 2 for _ in range(2)]
a[0][0] = 1
print(a)
gives the correct answer
[[1,0],[0,0]]
Related
for i in range(len(arr)):
temps = [[0]*9]*3
# temp = [[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]]
when tries
temps[0][0] = 4
gives this output
[[4, 0, 0, 0, 0, 0, 0, 0, 0], [4, 0, 0, 0, 0, 0, 0,
0, 0], [4, 0, 0, 0, 0, 0, 0, 0, 0]]
tries to change the value of temp[0][0]
it changes all temp[0][0] temp[0][1] temp[0][2]
value
Christian is right. [t] * 3 is equivalent to [t, t, t]
You can solve it by doing this:
for i in range(len(arr)):
temps = [[0]*9 for _ in range(3)]
Assuming that amount is the amount of the array, and b is the length of an array. I have no idea how to fill this
def MultiList(amount,length)
I want to if i call MultiList function like
MultiList(5,5)
Yhe output will be
(array ([0,0,0,0,0]), array ([0,0,0,0,0]), array ([0,0,0,0,0]), array ([0,0,0,0,0]),array ([0,0,0,0,0]))
For your simple case:
def gen_multi_list(amount, length, value=0):
return [[value]*length for _ in range(amount)]
print(gen_multi_list(5,5))
The 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]]
You can use numpy tuple to do this task.
import numpy as np
def multilist(amount, length):
return tuple(np.zeros(length, dtype=np.int) for _ in range(amount))
print(multilist(5,5))
Output :
(array([0, 0, 0, 0, 0]), array([0, 0, 0, 0, 0]), array([0, 0, 0, 0, 0]), array([0, 0, 0, 0, 0]), array([0, 0, 0, 0, 0]))
You can create the multiple array just using consecutive loops in function. to create list of list then convert it into tuple as we want tuple like output.
def Multilist(amount, length):
tup = [];
for i in range(amount):
arr = []
for j in range(length):
arr.append(0)
tup.append(arr)
return tuple(tup)
print(Multilist(5,5))
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])
Multilist = lambda amount, length : tuple([[0]*amount ]*length)
print(Multilist(5,5))
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])
mydata is an numpy array of shape(10,100,100) of the form(z,y,x). And i have created the empty array of shape(10,800,800). Now i need to place the mydata_array into some random locations of empty_array such that if I would plot the output, it should look like mydata is placed randomly in the ouput plot of array(10,800,800).
I used the np.hstack() and np.vstack().
But it places the mydata_array side by side. I need to place my_data_array in random location.
How could i do this? Any Suggestions please..
Regards
Raj
Here's a demonstration of placing several copies of one array inside another, using slice indexing:
In [802]: out = np.zeros((10,10),int)
In [803]: src = np.arange(6).reshape(2,3)
In [804]: out
Out[804]:
array([[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]])
One copy in the upper left:
In [805]: out[:2,:3] = src
In [806]: out
Out[806]:
array([[0, 1, 2, 0, 0, 0, 0, 0, 0, 0],
[3, 4, 5, 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]])
Several more copies:
In [808]: out[4:6, 6:9] = src
In [809]: out[1:3, 4:7] = src
In [810]: out
Out[810]:
array([[0, 1, 2, 0, 0, 0, 0, 0, 0, 0],
[3, 4, 5, 0, 0, 1, 2, 0, 0, 0],
[0, 0, 0, 0, 3, 4, 5, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 1, 2, 0],
[0, 0, 0, 0, 0, 0, 3, 4, 5, 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]])
Just repeat that kind of action for a selection of random locations. Make sure that the slice ranges match the src shape, and that they lie within the dimensions of the target array.
While may be possible to insert many copies at once (the flattening of the answer may be needed), let's start with understanding how to insert one copy at a time.
=========
#alvis' answer places the src items in shuffled order on one row of the out (or wrapped rows):
array([[2, 4, 5, 3, 0, 1, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
...
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0]])
===================
Looped placement of multiple blocks:
def foo1(src, idx, NM):
out = np.zeros(NM, dtype=src.dtype)
n,m = src.shape
for i,j in idx:
out[i:i+n, j:j+m] = src
return out
idx=np.array([[0,0],[1,4],[4,4],[8,7],[7,2]])
In [940]: out1 = foo1(src, idx, (10,10))
In [941]: out1
Out[941]:
array([[0, 1, 2, 0, 0, 0, 0, 0, 0, 0],
[3, 4, 5, 0, 0, 1, 2, 0, 0, 0],
[0, 0, 0, 0, 3, 4, 5, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 1, 2, 0, 0, 0],
[0, 0, 0, 0, 3, 4, 5, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 1, 2, 0, 0, 0, 0, 0],
[0, 0, 3, 4, 5, 0, 0, 0, 1, 2],
[0, 0, 0, 0, 0, 0, 0, 3, 4, 5]])
================
Placement of a block with advanced indexing (arrays instead of slices):
In [880]: I = np.array([1,1,1,2,2,2])
In [881]: J = np.array([3,4,5,3,4,5])
In [882]: out[I,J] = src.flat
In [883]: out
Out[883]:
array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 1, 2, 0, 0, 0, 0],
[0, 0, 0, 3, 4, 5, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
...
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0]])
And for multiple blocks
def foo2(src, idx, NM):
out = np.zeros(NM, dtype=src.dtype)
n,m = src.shape
ni = len(idx)
IJ = [np.mgrid[i:i+n, j:j+m] for i,j in idx]
IJ = np.concatenate(IJ, axis=1).reshape(2,-1)
out[IJ[0,:], IJ[1,:]] = np.tile(src,(ni,1)).flat
return out
In this small example the alternate is considerably slower (14x). For (1000,1000) out it is still slow (6x). Most of the time is spent in generating IJ.
This handles the I,J index calculation much faster (it needs to be generalize), but it is still slower than the looped slicing:
def foo3(src, idx, NM):
out = np.zeros(NM, dtype=src.dtype)
n,m = src.shape
ni = len(idx)
I = np.repeat((idx[:,[0]]+np.arange(2)).flatten(),3)
J = np.repeat((idx[:,[1]]+np.arange(3)),2,axis=0).flatten()
out[I, J] = np.tile(src,(ni,1)).flat
return out
This reminds me of work I did years ago to speed up the creation of a finite element stiffness matrix in MATLAB. There it was per-element stiffness blocks that needed to be placed in a large sparse global stiffness matrix.
==================
Regular pattern with broadcasting (see edit history)
According to your question, you don't need to preserve elements relatively to the first dimension of your array. For example, if there is one non-zero element a in (100,100) matrix z=0, and two elements b and c in the matrix z=1, then in your output all a, b, c can appear in z=0. In this case I suggest the following solution:
import numpy as np
#replace this with your input data
mydata = np.ones((10,100,100))
mydata_large = np.zeros((10,800,800))
mydata_flatten = mydata.flatten()
ind = np.array([i for i in range(len(mydata_flatten))])
np.random.shuffle(ind)
mydata_large_f = mydata_large.flatten()
np.put(mydata_large_f,ind[:len(mydata_flatten)],mydata_flatten)
mydata_large = np.reshape(mydata_large_f, (10,800,800))
I have an column vector that signifies the day of the week
[1,2,2,3,4]
I need to binarise this vector in the sense that every item in the original vector must be transformed to a vector where the number indicates an index that needs to be 1 and the rest must be 0.
[[0,1,0,0,0,0,0,0,0],
[0,0,1,0,0,0,0,0,0],
[0,0,1,0,0,0,0,0,0],
[0,0,0,1,0,0,0,0,0],
[0,0,0,0,1,0,0,0,0]]
do it by composing your binary list with zeroes except in the given position in a list comprehension which gives a nice one-liner:
w=[1,2,2,3,4]
m = [[0]*(pos)+[1]+[0]*(9-pos-1) for pos in w]
result:
m = [[0, 1, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 1, 0, 0, 0, 0]]
A simple list comprehension would be:
>> vector = [1,2,2,3,4]
>> [[int(i==j) for i in range(10)] for j in vector]
[[0, 1, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 1, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 1, 0, 0, 0, 0, 0]]
How can i iterate through a list of lists so as to make any of the lists with a "1" have the top(0), top left(0), top right(0), bottom(0), bottom right(0),bottom left(0) also become a "1" as shown below? making list 1 become list 2
list_1 =[[0,0,0,0,0,0,0,0],
[0,0,0,0,0,0,0,0],
[0,0,0,1,0,0,0,0],
[0,0,0,0,0,0,0,0]]
list_2 =[[0,0,0,0,0,0,0,0],
[0,0,1,1,1,0,0,0],
[0,0,1,1,1,0,0,0],
[0,0,1,1,1,0,0,0]]
This is a common operation known as "dilation" in image processing. Your problem is 2-dimensional, so you would be best served using
a more appropriate 2-d data structure than a list of lists, and
an already available library function, rather than reinvent the wheel
Here is an example using a numpy ndarray and scipy's binary_dilation respectively:
>>> import numpy as np
>>> from scipy import ndimage
>>> a = np.array([[0,0,0,0,0,0,0,0],
[0,0,0,0,0,0,0,0],
[0,0,0,1,0,0,0,0],
[0,0,0,0,0,0,0,0]], dtype=int)
>>> ndimage.binary_dilation(a, structure=ndimage.generate_binary_structure(2, 2)).astype(a.dtype)
array([[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 1, 1, 0, 0, 0],
[0, 0, 1, 1, 1, 0, 0, 0],
[0, 0, 1, 1, 1, 0, 0, 0]])
With numpy, which is more suitable to manipulate 2D list in general. If you're doing image analysis, see #wim answer. Otherwise here is how you could manage it with numpy only.
> import numpy as np
> list_1 =[[0,0,0,0,0,0,0,0],
[0,0,0,0,0,0,0,0],
[0,0,0,1,0,0,0,0],
[0,0,0,0,0,0,0,0]]
> l = np.array(list_1) # convert the list into a numpy array
> pos = np.where(l==1) # get the position where the array is equal to one
> pos
(array([2]), array([3]))
# make a lambda function to limit the lower indexes:
get_low = lambda x: x-1 if x>0 else x
# get_high is not needed.
# slice the array around that position and set the value to one
> l[get_low(pos[0]):pos[0]+2,
get_low(pos[1]):pos[1]+2] = 1
> l
array([[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 1, 1, 0, 0, 0],
[0, 0, 1, 1, 1, 0, 0, 0],
[0, 0, 1, 1, 1, 0, 0, 0]])
> corner
array([[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, 1]])
> p = np.where(corner==1)
> corner[get_low(p[0]):p[0]+2,
get_low(p[1]):p[1]+2] = 1
> corner
array([[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 1, 1],
[0, 0, 0, 0, 0, 0, 1, 1]])
HTH