bin2int function in R package rnn appears to have a bug - r

Question

I'm aware of the previous post asking about this, but there was not an answer, only the suggestion to write your own functions. I have the same issue--the functions int2bin and bin2int in the rnn package in R appear to return incorrect values. The problem appears to be in bin2int. I would appreciate verification this is a bug.
library(rnn)
X2 <- 1:154
X21 <- int2bin(X2, length = 15)
> head(X2)
[1] 1 2 3 4 5 6
# X21 (data after int2bin(X2, length = 15)) num [1:154, 1:15] 1 0 1 0 1 1 1...
>head(X21)
[,1][,2][,3][,4][,5][,6][,7][,8][,9][,10][,11][,12][,13][,14][,15]
[1,] 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
[2,] 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
[3,] 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0
[4,] 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
[5,] 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0
[6,] 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0
# so far so good
>X22 <- bin2int(X21)
# X22 (data after conversion back to integer) X22 int [1:154] 131072 262144...
> head(X22)
[1] 131072 262144 393216 524288 655360 786432
# should be 1 2 3 4 5 6

Answer 1

The underlying function of int2bin is i2b which is defined as:
function (integer, length = 8)
{
as.numeric(intToBits(integer))[1:length]
}
Which is then wrapped in int2bin
function (integer, length = 8)
{
t(sapply(integer, i2b, length = length))
}
Which is wrong (I think) because it returns the binary number backwards.
In your example 1 is returned as 100000000000000, when it should be returned as 000000000000001.
You can fix that by returning the intToBits() list backwards by changing [1:length] to [length:1]
function (integer, length = 8)
{
as.numeric(intToBits(integer))[length:1]
}
However, there is also a problem with bin2int, passing the correct binary input still outputs nonsense.
The b2i function is implemented as:
function(binary){
packBits(as.raw(c(rep(0, 32 - length(binary)), binary)), "integer")
}
Passing sample inputs, I don't understand what this function is doing - certainly not converting binary to integer.
Borrowing a function to convert binary to decimal from #Julius:
BinToDec <- function(x){
sum(2^(which(rev(unlist(strsplit(as.character(x), "")) == 1))-1)) }
This is a simple conversion from base2. Splits each binary digit, returns the indices where == 1, subtract 1 from each index (because R indexes from 1, not zero), then raise 2 to the power of each index returned earlier and sum. For example 101 (binary) = 2^2 + 2^0 = 5
And then (note this is using a corrected X21 structure that follows standard right-to-left binary notation)
X22 <- apply(X21,1,BinToDec)
Returns 1:154
So in short, yes, I agree that rnn:bin2int and rnn::int2bin appear to be wrong/broken.
Also, rather than trying to fix the rnn::int2bin function, I'd suggest R.utils::intToBin
And simply use:
require(R.utils)
X99 <- sapply(X2, intToBin)

Answer 2

I found an example implementation using rnn here:
https://www.r-bloggers.com/plain-vanilla-recurrent-neural-networks-in-r-waves-prediction/
This implementation works. I found the key to be the transpose before using trainr:
# Create sequences
t <- seq(0.005,2,by=0.005)
x <- sin(t*w) + rnorm(200, 0, 0.25)
y <- cos(t*w)
# Samples of 20 time series
X <- matrix(x, nrow = 40)
Y <- matrix(y, nrow = 40)
# Standardize in the interval 0 - 1
X <- (X - min(X)) / (max(X) - min(X))
Y <- (Y - min(Y)) / (max(Y) - min(Y))
# Transpose
X <- t(X)
Y <- t(Y)
Updating my code I had success in using the package. Therefore, the issues with the use of binary, if any, are not impacting the use of the package and were probably a red herring raised by me as I was searching for why my code wasn't producing expected results.