numpy Matrix Multiplication Simplification - is it possible? - python

Is there a way to simplify
a=np.dot(a,b)
just like the way you write a=a+b as a+=b ? (a,b are both np.array)

In Python3.5+ you can use the # operator for matrix multiplication, e.g.:
import numpy as np
a = np.random.randn(4, 10)
b = np.random.randn(10, 5)
c = a # b
This is equivalent to calling c = np.matmul(a, b). Inplace matrix multiplication (#=) is not yet supported (and doesn't make sense in most cases anyway, since the output usually has different dimensions to the first input).
Also note that np.matmul (and #) will behave differently to np.dot when one or more of the input arrays has >2 dimensions (see here).

Related

How to vectorize multiple matrix multiplication

I have a 2d matrix A[1000*90] and B[90*90*1000]
I would like to calculate C[1000*90]
For i in range(1000)
C[i,:]=np.matmul(A[i,:],B[:,:,i]
I understand if I use a vectorized formula it's going to be faster, seems like einsum might be the function I'm looking for, but I am having trouble cyphering the syntax of einsum. Is it np.einsum(ij,jki->ik,A,B)?
Your einsum is correct. But there is a better way as pointed out by hpaulj.
Using Matmul:
import numpy as np
A =np.random.rand(1000,90)
B = np.random.rand(90,90,1000)
C = A[:,np.newaxis,:]#B.transpose(2,0,1) ## Matrix multiplication
C = C = C.reshape(-1,C.shape[2])
np.array_equal(C,np.einsum('ij,jki->ik',A,B)) # check if both give same result

Python: numpy.dot / numpy.tensordot for multidimensional arrays

I'm optimising my implementation of the back-propagation algorithm to train a neural network. One of the aspects I'm working on is performing the matrix operations on the set of datapoints (input/output vector) as a batch process optimised by the numpy library instead of looping through every datapoint.
In my original algorithm I did the following:
for datapoint in datapoints:
A = ... (created out of datapoint info)
B = ... (created out of datapoint info)
C = np.dot(A,B.transpose())
____________________
A: (7,1) numpy array
B: (6,1) numpy array
C: (7,6) numpy array
I then expanded said matrices to tensors, where the first shape index would refer to the dataset. If I have 3 datasets (for simplicity purposes), the matrices would look like this:
A: (3,7,1) numpy array
B: (3,6,1) numpy array
C: (3,7,6) numpy array
Using ONLY np.tensordot or other numpy manipulations, how do I generate C?
I assume the answer would look something like this:
C = np.tensordot(A.[some manipulation], B.[some manipulation], axes = (...))
(This is a part of a much more complex application, and the way I'm structuring things is not flexible anymore. If I find no solution I will only loop through the datasets and perform the multiplication for each dataset)
We can use np.einsum -
c = np.einsum('ijk,ilm->ijl',a,b)
Since the last axes are singleton, you might be better off with sliced arrays -
c = np.einsum('ij,il->ijl',a[...,0],b[...,0])
With np.matmul/#-operator -
c = a#b.swapaxes(1,2)

How to efficiently operate on sub-arrays like calculating the determinants, inverse,

I have to to multiple operations on sub-arrays like matrix inversions or building determinants. Since for-loops are not very fast in Python I wonder what is the best way to do this.
import numpy as np
n = 8
a = np.random.rand(3,3,n)
b = np.empty(n)
c = np.zeros_like(a)
for i in range(n):
b[i] = np.linalg.det(a[:,:,i])
c[:,:,i] = np.linalg.inv(a[:,:,i])
Those numpy.linalg functions would accept n-dim arrays as long as the last two axes are the ones that form the 2D slices along which functions are intended to be operated upon. Hence, to solve our cases, permute axes to bring-up the axis of iteration as the first one, perform the required operation and if needed push-back that axis back to it's original place.
Hence, we could get those outputs, like so -
b = np.linalg.det(np.moveaxis(a,2,0))
c = np.moveaxis(np.linalg.inv(np.moveaxis(a,2,0)),0,2)

Does np.dot automatically transpose vectors?

I am trying to calculate the first and second order moments for a portfolio of stocks (i.e. expected return and standard deviation).
expected_returns_annual
Out[54]:
ticker
adj_close CNP 0.091859
F -0.007358
GE 0.095399
TSLA 0.204873
WMT -0.000943
dtype: float64
type(expected_returns_annual)
Out[55]: pandas.core.series.Series
weights = np.random.random(num_assets)
weights /= np.sum(weights)
returns = np.dot(expected_returns_annual, weights)
So normally the expected return is calculated by
(x1,...,xn' * (R1,...,Rn)
with x1,...,xn are weights with a constraint that all the weights have to sum up to 1 and ' means that the vector is transposed.
Now I am wondering a bit about the numpy dot function, because
returns = np.dot(expected_returns_annual, weights)
and
returns = np.dot(expected_returns_annual, weights.T)
give the same results.
I tested also the shape of weights.T and weights.
weights.shape
Out[58]: (5,)
weights.T.shape
Out[59]: (5,)
The shape of weights.T should be (,5) and not (5,), but numpy displays them as equal (I also tried np.transpose, but there is the same result)
Does anybody know why numpy behave this way? In my opinion the np.dot product automatically shape the vector the right why so that the vector product work well. Is that correct?
Best regards
Tom
The semantics of np.dot are not great
As Dominique Paul points out, np.dot has very heterogenous behavior depending on the shapes of the inputs. Adding to the confusion, as the OP points out in his question, given that weights is a 1D array, np.array_equal(weights, weights.T) is True (array_equal tests for equality of both value and shape).
Recommendation: use np.matmul or the equivalent # instead
If you are someone just starting out with Numpy, my advice to you would be to ditch np.dot completely. Don't use it in your code at all. Instead, use np.matmul, or the equivalent operator #. The behavior of # is more predictable than that of np.dot, while still being convenient to use. For example, you would get the same dot product for the two 1D arrays you have in your code like so:
returns = expected_returns_annual # weights
You can prove to yourself that this gives the same answer as np.dot with this assert:
assert expected_returns_annual # weights == expected_returns_annual.dot(weights)
Conceptually, # handles this case by promoting the two 1D arrays to appropriate 2D arrays (though the implementation doesn't necessarily do this). For example, if you have x with shape (N,) and y with shape (M,), if you do x # y the shapes will be promoted such that:
x.shape == (1, N)
y.shape == (M, 1)
Complete behavior of matmul/#
Here's what the docs have to say about matmul/# and the shapes of inputs/outputs:
If both arguments are 2-D they are multiplied like conventional matrices.
If either argument is N-D, N > 2, it is treated as a stack of matrices residing in the last two indexes and broadcast accordingly.
If the first argument is 1-D, it is promoted to a matrix by prepending a 1 to its dimensions. After matrix multiplication the prepended 1 is removed.
If the second argument is 1-D, it is promoted to a matrix by appending a 1 to its dimensions. After matrix multiplication the appended 1 is removed.
Notes: the arguments for using # over dot
As hpaulj points out in the comments, np.array_equal(x.dot(y), x # y) for all x and y that are 1D or 2D arrays. So why do I (and why should you) prefer #? I think the best argument for using # is that it helps to improve your code in small but significant ways:
# is explicitly a matrix multiplication operator. x # y will raise an error if y is a scalar, whereas dot will make the assumption that you actually just wanted elementwise multiplication. This can potentially result in a hard-to-localize bug in which dot silently returns a garbage result (I've personally run into that one). Thus, # allows you to be explicit about your own intent for the behavior of a line of code.
Because # is an operator, it has some nice short syntax for coercing various sequence types into arrays, without having to explicitly cast them. For example, [0,1,2] # np.arange(3) is valid syntax.
To be fair, while [0,1,2].dot(arr) is obviously not valid, np.dot([0,1,2], arr) is valid (though more verbose than using #).
When you do need to extend your code to deal with many matrix multiplications instead of just one, the ND cases for # are a conceptually straightforward generalization/vectorization of the lower-D cases.
I had the same question some time ago. It seems that when one of your matrices is one dimensional, then numpy will figure out automatically what you are trying to do.
The documentation for the dot function has a more specific explanation of the logic applied:
If both a and b are 1-D arrays, it is inner product of vectors
(without complex conjugation).
If both a and b are 2-D arrays, it is matrix multiplication, but using
matmul or a # b is preferred.
If either a or b is 0-D (scalar), it is equivalent to multiply and
using numpy.multiply(a, b) or a * b is preferred.
If a is an N-D array and b is a 1-D array, it is a sum product over
the last axis of a and b.
If a is an N-D array and b is an M-D array (where M>=2), it is a sum
product over the last axis of a and the second-to-last axis of b:
In NumPy, a transpose .T reverses the order of dimensions, which means that it doesn't do anything to your one-dimensional array weights.
This is a common source of confusion for people coming from Matlab, in which one-dimensional arrays do not exist. See Transposing a NumPy Array for some earlier discussion of this.
np.dot(x,y) has complicated behavior on higher-dimensional arrays, but its behavior when it's fed two one-dimensional arrays is very simple: it takes the inner product. If we wanted to get the equivalent result as a matrix product of a row and column instead, we'd have to write something like
np.asscalar(x # y[:, np.newaxis])
adding a trailing dimension to y to turn it into a "column", multiplying, and then converting our one-element array back into a scalar. But np.dot(x,y) is much faster and more efficient, so we just use that.
Edit: actually, this was dumb on my part. You can, of course, just write matrix multiplication x # y to get equivalent behavior to np.dot for one-dimensional arrays, as tel's excellent answer points out.
The shape of weights.T should be (,5) and not (5,),
suggests some confusion over the shape attribute. shape is an ordinary Python tuple, i.e. just a set of numbers, one for each dimension of the array. That's analogous to the size of a MATLAB matrix.
(5,) is just the way of displaying a 1 element tuple. The , is required because of older Python history of using () as a simple grouping.
In [22]: tuple([5])
Out[22]: (5,)
Thus the , in (5,) does not have a special numpy meaning, and
In [23]: (,5)
File "<ipython-input-23-08574acbf5a7>", line 1
(,5)
^
SyntaxError: invalid syntax
A key difference between numpy and MATLAB is that arrays can have any number of dimensions (upto 32). MATLAB has a lower boundary of 2.
The result is that a 5 element numpy array can have shapes (5,), (1,5), (5,1), (1,5,1)`, etc.
The handling of a 1d weight array in your example is best explained the np.dot documentation. Describing it as inner product seems clear enough to me. But I'm also happy with the
sum product over the last axis of a and the second-to-last axis of b
description, adjusted for the case where b has only one axis.
(5,) with (5,n) => (n,) # 5 is the common dimension
(n,5) with (5,) => (n,)
(n,5) with (5,1) => (n,1)
In:
(x1,...,xn' * (R1,...,Rn)
are you missing a )?
(x1,...,xn)' * (R1,...,Rn)
And the * means matrix product? Not elementwise product (.* in MATLAB)? (R1,...,Rn) would have size (n,1). (x1,...,xn)' size (1,n). The product (1,1).
By the way, that raises another difference. MATLAB expands dimensions to the right (n,1,1...). numpy expands them to the left (1,1,n) (if needed by broadcasting). The initial dimensions are the outermost ones. That's not as critical a difference as the lower size 2 boundary, but shouldn't be ignored.

find 2d elements in a 3d array which are similar to 2d elements in another 3d array

I have two 3D arrays and want to identify 2D elements in one array, which have one or more similar counterparts in the other array.
This works in Python 3:
import numpy as np
import random
np.random.seed(123)
A = np.round(np.random.rand(25000,2,2),2)
B = np.round(np.random.rand(25000,2,2),2)
a_index = np.zeros(A.shape[0])
for a in range(A.shape[0]):
for b in range(B.shape[0]):
if np.allclose(A[a,:,:].reshape(-1, A.shape[1]), B[b,:,:].reshape(-1, B.shape[1]),
rtol=1e-04, atol=1e-06):
a_index[a] = 1
break
np.nonzero(a_index)[0]
But of course this approach is awfully slow. Please tell me, that there is a more efficient way (and what it is). THX.
You are trying to do an all-nearest-neighbor type query. This is something that has special O(n log n) algorithms, I'm not aware of a python implementation. However you can use regular nearest-neighbor which is also O(n log n) just a bit slower. For example scipy.spatial.KDTree or cKDTree.
import numpy as np
import random
np.random.seed(123)
A = np.round(np.random.rand(25000,2,2),2)
B = np.round(np.random.rand(25000,2,2),2)
import scipy.spatial
tree = scipy.spatial.cKDTree(A.reshape(25000, 4))
results = tree.query_ball_point(B.reshape(25000, 4), r=1e-04, p=1)
print [r for r in results if r != []]
# [[14252], [1972], [7108], [13369], [23171]]
query_ball_point() is not an exact equivalent to allclose() but it is close enough, especially if you don't care about the rtol parameter to allclose(). You also get a choice of metric (p=1 for city block, or p=2 for Euclidean).
P.S. Consider using query_ball_tree() for very large data sets. Both A and B have to be indexed in that case.
P.S. I'm not sure what effect the 2d-ness of the elements should have; the sample code I gave treats them as 1d and that is identical at least when using city block metric.
From the docs of np.allclose, we have :
If the following equation is element-wise True, then allclose returns
True.
absolute(a - b) <= (atol + rtol * absolute(b))
Using that criteria, we can have a vectorized implementation using broadcasting, customized for the stated problem, like so -
# Setup parameters
rtol,atol = 1e-04, 1e-06
# Use np.allclose criteria to detect true/false across all pairwise elements
mask = np.abs(A[:,None,] - B) <= (atol + rtol * np.abs(B))
# Use the problem context to get final output
out = np.nonzero(mask.all(axis=(2,3)).any(1))[0]

Categories