Einsum for high dimensions - python

Considering the 3 arrays below:
np.random.seed(0)
X = np.random.randint(10, size=(4,5))
W = np.random.randint(10, size=(3,4))
y = np.random.randint(3, size=(5,1))
i want to add and sum each column of the matrix X to the row of W ,given by y as index. So ,for example, if the first element in y is 3 , i'll add the first column of X to the fourth row of W(index 3 in python) and sum it. i'll do it over and over until all columns of X are added to the specific row of W and summed.
i could do it in different ways:
1- using for loop:
for i,j in enumerate(y):
W[j]+=X[:,i]
2- using the add.at function
np.add.at(W,(y.ravel()),X.T)
3- but i can't understand how to do it using einsum.
i was given a solution ,but really can't understand it.
N = y.max()+1
W[:N] += np.einsum('ijk,lk->il',(np.arange(N)[:,None,None] == y.ravel()),X)
Anyone could explain me this structure?
1 - (np.arange(N)[:,None,None] == y.ravel(),X). i imagine this part refers to summing the column of X to the specific row of W ,according to y. But where s W ? and why do we have to transform W in 4 dimensions in this case?
2- 'ijk,lk->il' - i didnt understand this either.
i -refers to the rows,
j - columns,
k- each element,
l - what does 'l' refers too?.
if anyone can understand this and explain to me , i would really appreciate.
Thanks in advance.

Let's simplify the problem by dropping one dimension and using values that are easy to verify manually:
W = np.zeros(3, np.int)
y = np.array([0, 1, 1, 2, 2])
X = np.array([1, 2, 3, 4, 5])
Values in the vector W get added values from X by looking up with y:
for i, j in enumerate(y):
W[j] += X[i]
W is calculated as [1, 5, 9], (check quickly by hand).
Now, how could this code be vectorized? We can't do a simple W[y] += X[y] as y has duplicate values in it and the different sums would overwrite each other at indices 1 and 2.
What could be done is to broadcast the values into a new dimension of len(y) and then sum up over this newly created dimension.
N = W.shape[0]
select = (np.arange(N) == y[:, None]).astype(np.int)
Taking the index range of W ([0, 1, 2]), and setting the values where they match y to 1 in a new dimension, otherwise 0. select contains this array:
array([[1, 0, 0],
[0, 1, 0],
[0, 1, 0],
[0, 0, 1],
[0, 0, 1]])
It has len(y) == len(X) rows and len(W) columns and shows for every y/row, what index of W it contributes to.
Let's multiply X with this array, mult = select * X[:, None]:
array([[1, 0, 0],
[0, 2, 0],
[0, 3, 0],
[0, 0, 4],
[0, 0, 5]])
We have effectively spread out X into a new dimension, and sorted it in a way we can get it into shape W by summing over the newly created dimension. The sum over the rows is the vector we want to add to W:
sum_Xy = np.sum(mult, axis=0) # [1, 5, 9]
W += sum_Xy
The computation of select and mult can be combined with np.einsum:
# `select` has shape (len(y)==len(X), len(W)), or `yw`
# `X` has shape len(X)==len(y), or `y`
# we want something `len(W)`, or `w`, and to reduce the other dimension
sum_Xy = np.einsum("yw,y->w", select, X)
And that's it for the one-dimensional example. For the two-dimensional problem posed in the question it is exactly the same approach: introduce an additional dimension, broadcast the y indices, and then reduce the additional dimension with einsum.
If you internalize how every step works for the one-dimensional example, I'm sure you can work out how the code is doing it in two dimensions, as it is just a matter of getting the indices right (W rows, X columns).

Related

PyTorch's torch.as_strided with negative strides for making a Toeplitz matrix

I am writing a jury-rigged PyTorch version of scipy.linalg.toeplitz, which currently has the following form:
def toeplitz_torch(c, r=None):
c = torch.tensor(c).ravel()
if r is None:
r = torch.conj(c)
else:
r = torch.tensor(r).ravel()
# Flip c left to right.
idx = [i for i in range(c.size(0)-1, -1, -1)]
idx = torch.LongTensor(idx)
c = c.index_select(0, idx)
vals = torch.cat((c, r[1:]))
out_shp = len(c), len(r)
n = vals.stride(0)
return torch.as_strided(vals[len(c)-1:], size=out_shp, stride=(-n, n)).copy()
But torch.as_strided currently does not support negative strides. My function, therefore, throws the error:
RuntimeError: as_strided: Negative strides are not supported at the moment, got strides: [-1, 1].
My (perhaps incorrect) understanding of as_strided is that it inserts the values of the first argument into a new array whose size is specified by the second argument and it does so by linearly indexing those values in the original array and placing them at subscript-indexed strides given by the final argument.
Both the NumPy and PyTorch documentation concerning as_strided have scary warnings about using the function with "extreme care" and I don't understand this function fully, so I'd like to ask:
Is my understanding of as_strided correct?
Is there a simple way to rewrite this so negative strides work?
Will I be able to pass a gradient w.r.t c (or r) through toeplitz_torch?
> 1. Is my understanding of as_strided correct?
The stride is an interface for your tensor to access the underlying contiguous data buffer. It does not insert values, no copies of the values are done by torch.as_strided, the strides define the artificial layout of what we refer to as multi-dimensional array (in NumPy) or tensor (in PyTorch).
As Andreas K. puts it in another answer:
Strides are the number of bytes to jump over in the memory in order to get from one item to the next item along each direction/dimension of the array. In other words, it's the byte-separation between consecutive items for each dimension.
Please feel free to read the answers over there if you have some trouble with strides. Here we will take your example and look at how it is implemented with as_strided.
The example given by Scipy for linalg.toeplitz is the following:
>>> toeplitz([1,2,3], [1,4,5,6])
array([[1, 4, 5, 6],
[2, 1, 4, 5],
[3, 2, 1, 4]])
To do so they first construct the list of values (what we can refer to as the underlying values, not actually underlying data): vals which is constructed as [3 2 1 4 5 6], i.e. the Toeplitz column and row flattened.
Now notice the arguments passed to np.lib.stride_tricks.as_strided:
values: vals[len(c)-1:] notice the slice: the tensors show up smaller, yet the underlying values remain, and they correspond to those of vals. Go ahead and compare the two with storage_offset: it's just an offset of 2, the values are still there! How this works is that it essentially shifts the indices such that index=0 will refer to value 1, index=1 to 4, etc...
shape: given by the column/row inputs, here (3, 4). This is the shape of the resulting object.
strides: this is the most important piece: (-n, n), in this case (-1, 1)
The most intuitive thing to do with strides is to describe a mapping between the multi-dimensional space: (i, j) ∈ [0,3[ x [0,4[ and the flattened 1D space: k ∈ [0, 3*4[. Since the strides are equal to (-n, n) = (-1, 1), the mapping is -n*i + n*j = -1*i + 1*j = j-i. Mathematically you can describe your matrix as M[i, j] = F[j-i] where F is the flattened values vector [3 2 1 4 5 6].
For instance, let's try with i=1 and j=2. If you look at the Topleitz matrix above M[1, 2] = 4. Indeed F[k] = F[j-i] = F[1] = 4
If you look closely you will see the trick behind negative strides: they allow you to 'reference' to negative indices: for instance, if you take j=0 and i=2, then you see k=-2. Remember how vals was given with an offset of 2 by slicing vals[len(c)-1:]. If you look at its own underlying data storage it's still [3 2 1 4 5 6], but has an offset. The mapping for vals (in this case i: 1D -> k: 1D) would be M'[i] = F'[k] = F'[i+2] because of the offset. This means M'[-2] = F'[0] = 3.
In the above I defined M' as vals[len(c)-1:] which basically equivalent to the following tensor:
>>> torch.as_strided(vals, size=(len(vals)-2,), stride=(1,), storage_offset=2)
tensor([1, 4, 5, 6])
Similarly, I defined F' as the flattened vector of underlying values: [3 2 1 4 5 6].
The usage of strides is indeed a very clever way to define a Toeplitz matrix!
> 2. Is there a simple way to rewrite this so negative strides work?
The issue is, negative strides are not implemented in PyTorch... I don't believe there is a way around it with torch.as_strided, otherwise it would be rather easy to extend the current implementation and provide support for that feature.
There are however alternative ways to solve the problem. It is entirely possible to construct a Toeplitz matrix in PyTorch, but that won't be with torch.as_strided.
We will do the mapping ourselves: for each element of M indexed by (i, j), we will find out the corresponding index k which is simply j-i. This can be done with ease, first by gathering all (i, j) pairs from M:
>>> i, j = torch.ones(3, 4).nonzero().T
(tensor([0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2]),
tensor([0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3]))
Now we essentially have k:
>>> j-i
tensor([ 0, 1, 2, 3, -1, 0, 1, 2, -2, -1, 0, 1])
We just need to construct a flattened tensor of all possible values from the row r and column c inputs. Negative indexed values (the content of c) are put last and flipped:
>>> values = torch.cat((r, c[1:].flip(0)))
tensor([1, 4, 5, 6, 3, 2])
Finally index values with k and reshape:
>>> values[j-i].reshape(3, 4)
tensor([[1, 4, 5, 6],
[2, 1, 4, 5],
[3, 2, 1, 4]])
To sum it up, my proposed implementation would be:
def toeplitz(c, r):
vals = torch.cat((r, c[1:].flip(0)))
shape = len(c), len(r)
i, j = torch.ones(*shape).nonzero().T
return vals[j-i].reshape(*shape)
> 3. Will I be able to pass a gradient w.r.t c (or r) through toeplitz_torch?
That's an interesting question because torch.as_strided doesn't have a backward function implemented. This means you wouldn't have been able to backpropagate to c and r! With the above method, however, which uses 'backward-compatible' builtins, the backward pass comes free of charge.
Notice the grad_fn on the output:
>>> toeplitz(torch.tensor([1.,2.,3.], requires_grad=True),
torch.tensor([1.,4.,5.,6.], requires_grad=True))
tensor([[1., 4., 5., 6.],
[2., 1., 4., 5.],
[3., 2., 1., 4.]], grad_fn=<ViewBackward>)
This was a quick draft (that did take a little while to write down), I will make some edits. If you have some questions or remarks, don't hesitate to comment! I would be interested in seeing other answers as I am not an expert with strides, this is just my take on the problem.

Replacing array at i`th dimension

Let's say I have a two-dimensional array
import numpy as np
a = np.array([[1, 1, 1], [2,2,2], [3,3,3]])
and I would like to replace the third vector (in the second dimension) with zeros. I would do
a[:, 2] = np.array([0, 0, 0])
But what if I would like to be able to do that programmatically? I mean, let's say that variable x = 1 contained the dimension on which I wanted to do the replacing. How would the function replace(arr, dimension, value, arr_to_be_replaced) have to look if I wanted to call it as replace(a, x, 2, np.array([0, 0, 0])?
numpy has a similar function, insert. However, it doesn't replace at dimension i, it returns a copy with an additional vector.
All solutions are welcome, but I do prefer a solution that doesn't recreate the array as to save memory.
arr[:, 1]
is basically shorthand for
arr[(slice(None), 1)]
that is, a tuple with slice elements and integers.
Knowing that, you can construct a tuple of slice objects manually, adjust the values depending on an axis parameter and use that as your index. So for
import numpy as np
arr = np.array([[1, 1, 1], [2, 2, 2], [3, 3, 3]])
axis = 1
idx = 2
arr[:, idx] = np.array([0, 0, 0])
# ^- axis position
you can use
slices = [slice(None)] * arr.ndim
slices[axis] = idx
arr[tuple(slices)] = np.array([0, 0, 0])

RuntimeError when changing the values of specific parts of a `torch.Tensor`

Say I have a 3 dimentional tensor x initialized with zeros:
x = torch.zeros((2, 2, 2))
and an other 3 dimentional tensor y
y = torch.ones((2, 1, 2))
I am trying to change the values of the first line of x[0] and x[1] like this
x[:, 0, :] = y
but I get this error:
RuntimeError: expand(torch.FloatTensor{[2, 1, 2]}, size=[2, 2]): the number of sizes provided (2) must be greater or equal to the number of dimensions in the tensor (3)
It is as if the tensor y was getting squeezed somehow. Is there a way around this?
Is this what you want?
x = torch.arange(0, 8).reshape((2,2,2))
y = torch.ones((2,2))
x2 = x.permute(1,0,2)
x2[0] = y
x_target = x2.permute(1,0,2)
The value of first rows of x are changed by y .
I found a straight forward way to do it:
x[:, 0, :] = y[:, 0, :]

build matrix from blocks

I have an object which is described by two quantities, A and B (in real case they can be more than two). Objects are correlated depending on the value of A and B. In particular I know the correlation matrix for A and for B. Just as example:
a = np.array([[1, 1, 0, 0],
[1, 1, 0, 0],
[0, 0, 1, 1],
[0, 0, 1, 1]])
b = np.array([[1, 1, 0],
[1, 1, 1],
[0, 1, 1]])
na = a.shape[0]
nb = b.shape[0]
correlation for A:
so if an element has A == 0.5 and the other equal to A == 1.5 they are fully correlated (red). Otherwise if an element has A == 0.5 and the second item has A == 3.5 they are uncorrelated (blue).
Similarly for B:
Now I want multiply the two correlation matrixes, but I want to obtain as final matrix a matrix with two axis, where the new axes are a folded version of the original axes:
def get_folded_bin(ia, ib):
return ia * nb + ib
here what I am doing:
result = np.swapaxes(np.tensordot(a, b, axes=0), 1, 2).reshape(na* nb, na * nb)
visually:
and in particular this must hold:
for ia1 in xrange(na):
for ia2 in xrange(na):
for ib1 in xrange(nb):
for ib2 in xrange(nb):
assert(a[ia1, ia2] * b[ib1, ib2] == result[get_folded_bin(ia1, ib1), get_folded_bin(ia2, ib2)])
actually my problem is to do it with more quantities (A, B, C, ...) in a general way. Maybe there is also a simpler function within numpy to do that.
np.einsum lets you simplify the tensordot expression a bit:
result = np.einsum('ij,kl->ikjl',a,b).reshape(-1, na * nb)
I don't think there's a way of eliminating the reshape.
It may also be easier to generalize to more arrays, though I wouldn't get carried away with too many iteration variables in one einsum expression.
I think finally I have found a solution:
np.kron(a,b)
and then I can compose with
np.kron(np.kron(a,b), c)

Roll rows of a matrix independently

I have a matrix (2d numpy ndarray, to be precise):
A = np.array([[4, 0, 0],
[1, 2, 3],
[0, 0, 5]])
And I want to roll each row of A independently, according to roll values in another array:
r = np.array([2, 0, -1])
That is, I want to do this:
print np.array([np.roll(row, x) for row,x in zip(A, r)])
[[0 0 4]
[1 2 3]
[0 5 0]]
Is there a way to do this efficiently? Perhaps using fancy indexing tricks?
Sure you can do it using advanced indexing, whether it is the fastest way probably depends on your array size (if your rows are large it may not be):
rows, column_indices = np.ogrid[:A.shape[0], :A.shape[1]]
# Use always a negative shift, so that column_indices are valid.
# (could also use module operation)
r[r < 0] += A.shape[1]
column_indices = column_indices - r[:, np.newaxis]
result = A[rows, column_indices]
numpy.lib.stride_tricks.as_strided stricks (abbrev pun intended) again!
Speaking of fancy indexing tricks, there's the infamous - np.lib.stride_tricks.as_strided. The idea/trick would be to get a sliced portion starting from the first column until the second last one and concatenate at the end. This ensures that we can stride in the forward direction as needed to leverage np.lib.stride_tricks.as_strided and thus avoid the need of actually rolling back. That's the whole idea!
Now, in terms of actual implementation we would use scikit-image's view_as_windows to elegantly use np.lib.stride_tricks.as_strided under the hoods. Thus, the final implementation would be -
from skimage.util.shape import view_as_windows as viewW
def strided_indexing_roll(a, r):
# Concatenate with sliced to cover all rolls
a_ext = np.concatenate((a,a[:,:-1]),axis=1)
# Get sliding windows; use advanced-indexing to select appropriate ones
n = a.shape[1]
return viewW(a_ext,(1,n))[np.arange(len(r)), (n-r)%n,0]
Here's a sample run -
In [327]: A = np.array([[4, 0, 0],
...: [1, 2, 3],
...: [0, 0, 5]])
In [328]: r = np.array([2, 0, -1])
In [329]: strided_indexing_roll(A, r)
Out[329]:
array([[0, 0, 4],
[1, 2, 3],
[0, 5, 0]])
Benchmarking
# #seberg's solution
def advindexing_roll(A, r):
rows, column_indices = np.ogrid[:A.shape[0], :A.shape[1]]
r[r < 0] += A.shape[1]
column_indices = column_indices - r[:,np.newaxis]
return A[rows, column_indices]
Let's do some benchmarking on an array with large number of rows and columns -
In [324]: np.random.seed(0)
...: a = np.random.rand(10000,1000)
...: r = np.random.randint(-1000,1000,(10000))
# #seberg's solution
In [325]: %timeit advindexing_roll(a, r)
10 loops, best of 3: 71.3 ms per loop
# Solution from this post
In [326]: %timeit strided_indexing_roll(a, r)
10 loops, best of 3: 44 ms per loop
In case you want more general solution (dealing with any shape and with any axis), I modified #seberg's solution:
def indep_roll(arr, shifts, axis=1):
"""Apply an independent roll for each dimensions of a single axis.
Parameters
----------
arr : np.ndarray
Array of any shape.
shifts : np.ndarray
How many shifting to use for each dimension. Shape: `(arr.shape[axis],)`.
axis : int
Axis along which elements are shifted.
"""
arr = np.swapaxes(arr,axis,-1)
all_idcs = np.ogrid[[slice(0,n) for n in arr.shape]]
# Convert to a positive shift
shifts[shifts < 0] += arr.shape[-1]
all_idcs[-1] = all_idcs[-1] - shifts[:, np.newaxis]
result = arr[tuple(all_idcs)]
arr = np.swapaxes(result,-1,axis)
return arr
I implement a pure numpy.lib.stride_tricks.as_strided solution as follows
from numpy.lib.stride_tricks import as_strided
def custom_roll(arr, r_tup):
m = np.asarray(r_tup)
arr_roll = arr[:, [*range(arr.shape[1]),*range(arr.shape[1]-1)]].copy() #need `copy`
strd_0, strd_1 = arr_roll.strides
n = arr.shape[1]
result = as_strided(arr_roll, (*arr.shape, n), (strd_0 ,strd_1, strd_1))
return result[np.arange(arr.shape[0]), (n-m)%n]
A = np.array([[4, 0, 0],
[1, 2, 3],
[0, 0, 5]])
r = np.array([2, 0, -1])
out = custom_roll(A, r)
Out[789]:
array([[0, 0, 4],
[1, 2, 3],
[0, 5, 0]])
By using a fast fourrier transform we can apply a transformation in the frequency domain and then use the inverse fast fourrier transform to obtain the row shift.
So this is a pure numpy solution that take only one line:
import numpy as np
from numpy.fft import fft, ifft
# The row shift function using the fast fourrier transform
# rshift(A,r) where A is a 2D array, r the row shift vector
def rshift(A,r):
return np.real(ifft(fft(A,axis=1)*np.exp(2*1j*np.pi/A.shape[1]*r[:,None]*np.r_[0:A.shape[1]][None,:]),axis=1).round())
This will apply a left shift, but we can simply negate the exponential exponant to turn the function into a right shift function:
ifft(fft(...)*np.exp(-2*1j...)
It can be used like that:
# Example:
A = np.array([[1,2,3,4],
[1,2,3,4],
[1,2,3,4]])
r = np.array([1,-1,3])
print(rshift(A,r))
Building on divakar's excellent answer, you can apply this logic to 3D array easily (which was the problematic that brought me here in the first place). Here's an example - basically flatten your data, roll it & reshape it after::
def applyroll_30(cube, threshold=25, offset=500):
flattened_cube = cube.copy().reshape(cube.shape[0]*cube.shape[1], cube.shape[2])
roll_matrix = calc_roll_matrix_flattened(flattened_cube, threshold, offset)
rolled_cube = strided_indexing_roll(flattened_cube, roll_matrix, cube_shape=cube.shape)
rolled_cube = triggered_cube.reshape(cube.shape[0], cube.shape[1], cube.shape[2])
return rolled_cube
def calc_roll_matrix_flattened(cube_flattened, threshold, offset):
""" Calculates the number of position along time axis we need to shift
elements in order to trig the data.
We return a 1D numpy array of shape (X*Y, time) elements
"""
# armax(...) finds the position in the cube (3d) where we are above threshold
roll_matrix = np.argmax(cube_flattened > threshold, axis=1) + offset
# ensure we don't have index out of bound
roll_matrix[roll_matrix>cube_flattened.shape[1]] = cube_flattened.shape[1]
return roll_matrix
def strided_indexing_roll(cube_flattened, roll_matrix_flattened, cube_shape):
# Concatenate with sliced to cover all rolls
# otherwise we shift in the wrong direction for my application
roll_matrix_flattened = -1 * roll_matrix_flattened
a_ext = np.concatenate((cube_flattened, cube_flattened[:, :-1]), axis=1)
# Get sliding windows; use advanced-indexing to select appropriate ones
n = cube_flattened.shape[1]
result = viewW(a_ext,(1,n))[np.arange(len(roll_matrix_flattened)), (n - roll_matrix_flattened) % n, 0]
result = result.reshape(cube_shape)
return result
Divakar's answer doesn't do justice to how much more efficient this is on large cube of data. I've timed it on a 400x400x2000 data formatted as int8. An equivalent for-loop does ~5.5seconds, Seberg's answer ~3.0seconds and strided_indexing.... ~0.5second.

Categories