Develop Reference

Numpy - Fastest way to create a integer index matrix

Problem:
I have an array that represents products, lets say 3 for example
prod = [1,2,3]
then I have a correlation matrix for those products (just a number that represents something between two products, lets call c_ij for simplicity), in this case a 3x3 matrix
corr = [[c_11,c_12,c_13],
[c_21,c_22,c_23],
[c_31,c_32,c_33]]
The problem is that a need to shuffle the prod array, then I need to shuffle the corr matrix in a way that corr[i,j] still represent the correlation between prod[i] and prod[j]
My solution:
I know I can use a integer array as index to shuffle multiple array in the same way, like this:
order = [0,1,2]
new_order = np.random.permutation(order) # [2,0,1] for example
shuf_prod = prod[new_order]
Looking in the web I find that to make this work in a matrix I need to transform the order array in a matrix like
new_order = [2,0,1]
new_index = [ [[2,2,2],[0,0,0],[1,1,1]],
[[2,0,1],[2,0,1],[2,0,1]] ]
new_corr = corr[tuple(new_index)]
# this output what I want that is:
# [[c_33,c_31,c_32],
# [c_13,c_11,c_12],
# [c_23,c_21,c_22]]
Question:
The entire solution of shuffling look chunky and not efficient, this is a performance critical application so there is a faster way to do this? (I don't really care for simplicity of code, just performance)
If this is a good way of doing this, how I can create the new_index matrix from new_order array?
EDIT: Michael Szczesny solved the problem
new_corr = corr[new_order].T[new_order].T

you can use the indices directly as subscripts to the matrix as long as you provide the right shape for the second axis:
import numpy as np
mat = np.array([[3,4,5],
[4,8,9],
[5,9,7]])
order = np.array([2,0,1])
mat[order,order[:,None]]
array([[7, 5, 9],
[5, 3, 4],
[9, 4, 8]])

How to use numpy to calculate mean and standard deviation of an irregular shaped array

I have a numpy array that has many samples in it of varying length
Samples = np.array([[1001, 1002, 1003],
... ,
[1001, 1002]])
I want to (elementwise) subtract the mean of the array then divide by the standard deviation of the array. Something like:
newSamples = (Samples-np.mean(Samples))/np.std(Samples)
Except that doesn't work for irregular shaped arrays,
np.mean(Samples) causes
unsupported operand type(s) for /: 'list' and 'int'
due to what I assume to be it having set a static size for each axis and then when it encounters a different sized sample it can't handle it. What is an approach to solve this using numpy?
example input:
Sample = np.array([[1, 2, 3],
[1, 2]])
After subtracting by the mean and then dividing by standard deviation:
Sample = array([[-1.06904497, 0.26726124, 1.60356745],
[-1.06904497, 0.26726124]])

Don't make ragged arrays. Just don't. Numpy can't do much with them, and any code you might make for them will always be unreliable and slow because numpy doesn't work that way. It turns them into object dtypes:
Sample
array([[1, 2, 3], [1, 2]], dtype=object)
Which almost no numpy functions work on. In this case those objects are list objects, which makes your code even more confusing as you either have to switch between list and ndarray methods, or stick to list-safe numpy methods. This a recipe for disaster as anyone noodling around with the code later (even yourself if you forget) will be dancing in a minefield.
There's two things you can do with your data to make things work better:
First method is to index and flatten.
i = np.cumsum(np.array([len(x) for x in Sample]))
flat_sample = np.hstack(Sample)
This preserves the index of the end of each sample in i, while keeping the sample as a 1D array
The other method is to pad one dimension with np.nan and use nan-safe functions
m = np.array([len(x) for x in Sample]).max()
nan_sample = np.array([x + [np.nan] * (m - len(x)) for x in Sample])
So to do your calculations, you can use flat_sample and do similar to above:
new_flat_sample = (flat_sample - np.mean(flat_sample)) / np.std(flat_sample)
and use i to recreate your original array (or list of arrays, which I recommend:, see np.split).
new_list_sample = np.split(new_flat_sample, i[:-1])
[array([-1.06904497, 0.26726124, 1.60356745]),
array([-1.06904497, 0.26726124])]
Or use nan_sample, but you will need to replace np.mean and np.std with np.nanmean and np.nanstd
new_nan_sample = (nan_sample - np.nanmean(nan_sample)) / np.nanstd(nan_sample)
array([[-1.06904497, 0.26726124, 1.60356745],
[-1.06904497, 0.26726124, nan]])

#MichaelHackman (following the comment remark).
That's weird because when I compute the overall std and mean then apply it, I obtain different result (see code below).
import numpy as np
Samples = np.array([[1, 2, 3],
[1, 2]])
c = np.hstack(Samples) # Will gives [1,2,3,1,2]
mean, std = np.mean(c), np.std(c)
newSamples = np.asarray([(np.array(xi)-mean)/std for xi in Samples])
print newSamples
# [array([-1.06904497, 0.26726124, 1.60356745]), array([-1.06904497, 0.26726124])]
edit: Add np.asarray(), put mean,std computation outside loop following Imanol Luengo's excellent comments (Thanks!)

np.bincount for 1 line, vectorized multidimensional averaging

I am trying to vectorize an operation using numpy, which I use in a python script that I have profiled, and found this operation to be the bottleneck and so needs to be optimized since I will run it many times.
The operation is on a data set of two parts. First, a large set (n) of 1D vectors of different lengths (with maximum length, Lmax) whose elements are integers from 1 to maxvalue. The set of vectors is arranged in a 2D array, data, of size (num_samples,Lmax) with trailing elements in each row zeroed. The second part is a set of scalar floats, one associated with each vector, that I have a computed and which depend on its length and the integer-value at each position. The set of scalars is made into a 1D array, Y, of size num_samples.
The desired operation is to form the average of Y over the n samples, as a function of (value,position along length,length).
This entire operation can be vectorized in matlab with use of the accumarray function: by using 3 2D arrays of the same size as data, whose elements are the corresponding value, position, and length indices of the desired final array:
sz_Y = num_samples;
sz_len = Lmax
sz_pos = Lmax
sz_val = maxvalue
ind_len = repmat( 1:sz_len ,1 ,sz_samples);
ind_pos = repmat( 1:sz_pos ,sz_samples,1 );
ind_val = data
ind_Y = repmat((1:sz_Y)',1 ,Lmax );
copiedY=Y(ind_Y);
mask = data>0;
finalarr=accumarray({ind_val(mask),ind_pos(mask),ind_len(mask)},copiedY(mask), [sz_val sz_pos sz_len])/sz_val;
I was hoping to emulate this implementation with np.bincounts. However, np.bincounts differs to accumarray in two relevant ways:
both arguments must be of same 1D size, and
there is no option to choose the shape of the output array.
In the above usage of accumarray, the list of indices, {ind_val(mask),ind_pos(mask),ind_len(mask)}, is 1D cell array of 1x3 arrays used as index tuples, while in np.bincounts it must be 1D scalars as far as I understand. I expect np.ravel may be useful but am not sure how to use it here to do what I want. I am coming to python from matlab and some things do not translate directly, e.g. the colon operator which ravels in opposite order to ravel. So my question is how might I use np.bincount or any other numpy method to achieve an efficient python implementation of this operation.
EDIT: To avoid wasting time: for these multiD index problems with complicated index manipulation, is the recommend route to just use cython to implement the loops explicity?
EDIT2: Alternative Python implementation I just came up with.
Here is a heavy ram solution:
First precalculate:
Using index units for length (i.e., length 1 =0) make a 4D bool array, size (num_samples,Lmax+1,Lmax+1,maxvalue) , holding where the conditions are satisfied for each value in Y.
ALLcond=np.zeros((num_samples,Lmax+1,Lmax+1,maxvalue+1),dtype='bool')
for l in range(Lmax+1):
for i in range(Lmax+1):
for v in range(maxvalue+!):
ALLcond[:,l,i,v]=(data[:,i]==v) & (Lvec==l)`
Where Lvec=[len(row) for row in data]. Then get the indices for these using np.where and initialize a 4D float array into which you will assign the values of Y:
[indY,ind_len,ind_pos,ind_val]=np.where(ALLcond)
Yval=np.zeros(np.shape(ALLcond),dtype='float')
Now in the loop in which I have to perform the operation, I compute it with the two lines:
Yval[ind_Y,ind_len,ind_pos,ind_val]=Y[ind_Y]
Y_avg=sum(Yval)/num_samples
This gives a factor of 4 or so speed up over the direct loop implementation. I was expecting more. Perhaps, this is a more tangible implementation for Python heads to digest. Any faster suggestions are welcome :)

One way is to convert the 3 "indices" to a linear index and then apply bincount. Numpy's ravel_multi_index is essentially the same as MATLAB's sub2ind. So the ported code could be something like:
shape = (Lmax+1, Lmax+1, maxvalue+1)
posvec = np.arange(1, Lmax+1)
ind_len = np.tile(Lvec[:,None], [1, Lmax])
ind_pos = np.tile(posvec, [n, 1])
ind_val = data
Y_copied = np.tile(Y[:,None], [1, Lmax])
mask = posvec <= Lvec[:,None] # fill-value independent
lin_idx = np.ravel_multi_index((ind_len[mask], ind_pos[mask], ind_val[mask]), shape)
Y_avg = np.bincount(lin_idx, weights=Y_copied[mask], minlength=np.prod(shape)) / n
Y_avg.shape = shape
This is assuming data has shape (n, Lmax), Lvec is Numpy array, etc. You may need to adapt the code a little to get rid of off-by-one errors.
One could argue that the tile operations are not very efficient and not very "numpythonic". Something with broadcast_arrays could be nice, but I think I prefer this way:
shape = (Lmax+1, Lmax+1, maxvalue+1)
posvec = np.arange(1, Lmax+1)
len_idx = np.repeat(Lvec, Lvec)
pos_idx = np.broadcast_to(posvec, data.shape)[mask]
val_idx = data[mask]
Y_copied = np.repeat(Y, Lvec)
mask = posvec <= Lvec[:,None] # fill-value independent
lin_idx = np.ravel_multi_index((len_idx, pos_idx, val_idx), shape)
Y_avg = np.bincount(lin_idx, weights=Y_copied, minlength=np.prod(shape)) / n
Y_avg.shape = shape
Note broadcast_to was added in Numpy 1.10.0.

Vectorizing a numpy array call of varying indices

I have a 2D numpy array and a list of lists of indices for which I wish to compute the sum of the corresponding 1D vectors from the numpy array. This can be easily done through a for loop or via list comprehension, but I wonder if it's possible to vectorize it. With similar code I gain about 40x speedups from the vectorization.
Here's sample code:
import numpy as np
indices = [[1,2],[1,3],[2,0,3],[1]]
array_2d = np.array([[0.5, 1.5],[1.5,2.5],[2.5,3.5],[3.5,4.5]])
soln = [np.sum(array_2d[x], axis=-1) for x in indices]
(edit): Note that the indices are not (x,y) coordinates for array_2d, instead indices[0] = [1,2] represents the first and second vectors (rows) in array_2d. The number of elements of each list in indices can be variable.
This is what I would hope to be able to do:
vectorized_soln = np.sum(array_2d[indices[:]], axis=-1)
Does anybody know if there are any ways of achieving this?

First to all, I think you have a typo in the third element of indices...
The easy way to do that is building a sub_array with two arrays of indices:
i = np.array([1,1,2])
j = np.array([2,3,?])
sub_arr2d = array_2d[i,j]
and finally, you can take the sum of sub_arr2d...

Convert a list of 2D numpy arrays to one 3D numpy array?

I have a list of several hundred 10x10 arrays that I want to stack together into a single Nx10x10 array. At first I tried a simple
newarray = np.array(mylist)
But that returned with "ValueError: setting an array element with a sequence."
Then I found the online documentation for dstack(), which looked perfect: "...This is a simple way to stack 2D arrays (images) into a single 3D array for processing." Which is exactly what I'm trying to do. However,
newarray = np.dstack(mylist)
tells me "ValueError: array dimensions must agree except for d_0", which is odd because all my arrays are 10x10. I thought maybe the problem was that dstack() expects a tuple instead of a list, but
newarray = np.dstack(tuple(mylist))
produced the same result.
At this point I've spent about two hours searching here and elsewhere to find out what I'm doing wrong and/or how to go about this correctly. I've even tried converting my list of arrays into a list of lists of lists and then back into a 3D array, but that didn't work either (I ended up with lists of lists of arrays, followed by the "setting array element as sequence" error again).
Any help would be appreciated.

newarray = np.dstack(mylist)
should work. For example:
import numpy as np
# Here is a list of five 10x10 arrays:
x = [np.random.random((10,10)) for _ in range(5)]
y = np.dstack(x)
print(y.shape)
# (10, 10, 5)
# To get the shape to be Nx10x10, you could use rollaxis:
y = np.rollaxis(y,-1)
print(y.shape)
# (5, 10, 10)
np.dstack returns a new array. Thus, using np.dstack requires as much additional memory as the input arrays. If you are tight on memory, an alternative to np.dstack which requires less memory is to
allocate space for the final array first, and then pour the input arrays into it one at a time.
For example, if you had 58 arrays of shape (159459, 2380), then you could use
y = np.empty((159459, 2380, 58))
for i in range(58):
# instantiate the input arrays one at a time
x = np.random.random((159459, 2380))
# copy x into y
y[..., i] = x