I'm struggling with implementing RTransferEntropy from Python

Question

I've written an R script in RStudio for purposes of computing transfer entropy between two vectors, x and y, and wish to call it from Python and verify that the results are consistent to see if I'm doing things right:

In RStudio:

TE <- function(x,y) {
  library(RTransferEntropy)
  library(future)

  plan(multiprocess)
  set.seed(12345)
  shannon_te <-transfer_entropy(x,y,nboot=1000)
  result=shannon_te
  return(result)
  }

The script compiles without error. Now I test it:

n <- 2500
x <- rep(0, n + 200)
y <- rep(0, n + 200)
x[1] <- rnorm(1, 0, 1)
y[1] <- rnorm(1, 0, 1)

for (i in 2:(n + 200)) {
  x[i] <- 0.2 * x[i - 1] + rnorm(1, .2, 1)
  y[i] <- sqrt(abs(x[i - 1])) + rnorm(1, .2, 1)
}

x <- x[-(1:200)]
y <- y[-(1:200)]

TE(x,y) works fine and produces transfer entropy values and their corresponding p-values, which show x has a causal effect on y, but not vice versa.

Now I wish to call this script from within Python using a function named T:

import numpy as np
from rpy2.robjects.packages import importr
import rpy2.robjects as ro

importr('RTransferEntropy',lib_loc='/Library/Frameworks/R.framework/
Versions/3.6/Resources/library')
base = importr('base')
utils = importr('utils')

import rpy2.robjects.packages as rpackages
import rpy2.robjects

def T(x,y):
    r=ro.r
    r.source("TE.R")
    t=r.TE(x,y)
    return t

None of this yields any errors. I want to test this by defining the same vectors, x and y, in Python and call the Python function above to verify that x has a causal influence on y but not vice versa:

time=np.arange(10)
for t in time:
    x[t+1]=.2*x[t]+np.random.normal(1,.2,1)
    y[t+1]=ma.sqrt(abs(x[t]))+np.random.normal(1,.2,1)  
T(x,y)

This produces the error message below:

score 0 · Answer 1 · answered Oct 01 '19 at 22:04

0

This is the first time I see or hear about that error. Note that an RRuntimeError is an error raised by the embedded R and propagated to Python by rpy2.

In order to figure out what is happening here, I would first take the library imports out of the body of the function to make the example shorter and narrow down possible roots for the problem. One way would be import these while still in Python with:

rtransferentropy = importr('RTransferEntropy')
future = import('future')

answered Oct 01 '19 at 22:04

lgautier

11,363
29
42

So replace my original importr with rtransferentropy = importr('RTransferEntropy') future = import('future') ??? That gives an error message starting with "R[write to console]: Error: package or namespace load failed for ‘RTransferEntropy’ HOWEVER, importr('RTransferEntropy',lib_loc="/Users/fishbacp/anaconda3/envs/RStudio/lib/R library") works. But then calling the T defined above gives NotImplementedError: Conversion 'py2rpy' not defined for objects of type '' Sorry I can't figure out code formatting here. – fishbacp Oct 02 '19 at 20:03
1

@fishbacp You can only use code formatting if you update the question, I think. – lgautier Oct 02 '19 at 22:08

score 0 · Answer 2 · answered Oct 02 '19 at 23:39

I found that the following worked:

Delete:

importr('RTransferEntropy',lib_loc='/Library/Frameworks/R.framework/
Versions/3.6/Resources/library')

Replace with:

importr('RTransferEntropy', 
lib_loc="/Users/fishbacp/anaconda3/envs/RStudio/lib/R/library")

Add:

import rpy2.robjects.numpy2ri
rpy2.robjects.numpy2ri.activate()

I got the idea from Converting python objects for rpy2.

I'm struggling with implementing RTransferEntropy from Python

2 Answers2