const-ae / einsum Goto Github PK

Hi,

Thanks for the great package.
This is really useful when the data structure is complex.
Do you have a plan to submit this package to the CRAN repository?
I want to use this functionality by importing this package.

Regards,

Hi
Congrats for this package. I saw that was already available on CRAN.
I would to thank you for filling this gap in R as this function is really useful.
However, I spotted some different behavior in comparison with einsum() from numpy.
The following script (see bellow)example illustrate this discrepancy. I hope this can be helpful. However, this is maybe a total error of my part, So I'll appreciate any answer and explanation.

Kind regards

Takiy

########################################################################################

python

import numpy as np
mat=np.array([[1,2,3],[4,5,6],[7,8,9]])
print(mat)

##########################################

[[1 2 3]
 [4 5 6]
 [7 8 9]]

##########################################

1/

np.einsum('mp,mq->mq',mat,mat)
###########################################

array([[  6,  12,  18],
       [ 60,  75,  90],
       [168, 192, 216]])

###########################################
# identical to einsum() in R
###########################################

2/

np.einsum('mp,mq->mpq',mat,mat)
###########################################

array([[[ 1,  2,  3],
        [ 2,  4,  6],
        [ 3,  6,  9]],

       [[16, 20, 24],
        [20, 25, 30],
        [24, 30, 36]],

       [[49, 56, 63],
        [56, 64, 72],
        [63, 72, 81]]])

##########################################
# not identical to einsum() in R
###########################################

R

library(einsum)
mat=matrix(1:9,ncol=3,byrow=T)
mat

##########################################

     [,1] [,2] [,3]
[1,]    1    2    3
[2,]    4    5    6
[3,]    7    8    9

##########################################

1/

einsum('mp,mq->mq',mat,mat)
#########################################

     [,1] [,2] [,3]
[1,]    6   12   18
[2,]   60   75   90
[3,]  168  192  216

#########################################
# identical to einsum() in python
###########################################

2/

einsum('mp,mq->mpq',mat,mat)
##########################################

, , 1

     [,1] [,2] [,3]
[1,]    1    2    3
[2,]   16   20   24
[3,]   49   56   63

, , 2

     [,1] [,2] [,3]
[1,]    2    4    6
[2,]   20   25   30
[3,]   56   64   72

, , 3

     [,1] [,2] [,3]
[1,]    3    6    9
[2,]   24   30   36
[3,]   63   72   81

##########################################
# not identical to einsum() in python
###########################################
########################################################################################

Hi, Constantin

I found that CRAN einsum cannot use in implicit mode without explicit indicator (->).

It'll be fine, implicit mode is sometimes misleading.
However, CRAN einsum can actually be performed in different meaning like below.

For example, in Numpy's einsum, implicit mode just shows the whole data.

# Numpy (Python)
import numpy as np
data = np.random.rand(3, 4)
np.einsum('ij', data)
# array([[0.50142663, 0.845945  , 0.36696267, 0.00643327],
#        [0.14395742, 0.73767592, 0.81586026, 0.40644605],
#        [0.12538026, 0.92932431, 0.2329707 , 0.10398253]])

However, in CRAN einsum, it is interpreted as the summation over the indices i and j.

# einsum (R)
library("einsum")
data <- array(runif(3*4), dim=c(3,4))
einsum('ij', data)
# [1] 2.951262

Wouldn't it be more trouble-free to ban equation_strings containing "->"?
For example, how about including the following argument check?

if(length(grep("->", "equation_string")) == 0){
	stop("Please make sure that equation_string has explicit indicator (->)")
}