Is all of functionality now in base or other packages for Julia 0.4

[timema]

I am just starting to move some of my code from Julia 0.3 to the new 0.4 release. When I discovered NumericExtensions in 0.3 it dramatically sped up some of my code. I noticed that NumericExtensions does not work with 0.4 and then I saw multiple posts on the git site and Julia-Users suggesting it was just a stopgap until 0.4 came out and that some of the stuff in NumericExtensions was put in base or other packages like StatFuns.

After a bit of searching I was able to find one of those functions in Base and I looked at StatFuns but I could not find any others. My question is have all of the equivalent functionality been moved to base and/or other packages or just some of it? If not, is there a generally way with respect to changes in 0.4 to reproduce some of it without being an expert (particularly for functions like wsumfdiff)? Below are the functions I have used and what I have found in base and the functions or equivalent functions I have not found yet.

Here are the functions that I have been using:

Found:
sum(Abs2Fun(),x)
found equivalent with sumabs2 in base in reduce.jl

Not Sure:
sum(LogFun(),x)
I am not sure if sum(log(x)) is now the same or as fast as sum(LogFun(),x)

Cannot Find:
wsumfdiff(rcp!(x),Abs2Fun(),yperm,that)
cannot find the equivalent of rcp! which is equivalent to x<- 1./x
cannot find the equivalent of wsumfdiff or wsumsqdiff

I found sqL2dist(a,b) in Statsbase but I am not sure if you can or how to incorporate weights

I am sorry for the noise if the answers to this have already been posted somewhere that I have not found.

[timema]

I looked at the sqL2dist function in StatsBase and thought I could modify it to include a weight to make it equivalent to wsumsqdiff in NumericExtensions. Below is the original sqL2dist function from deviation.jl of StatsBase then a modification to incorporate weights and then another modification that incorporates the inverse of a vector for the weights (my use case). I would like to know if these look reasonable (i.e. as fast an efficient). I am not asking to merge these into Statsbase or anything like that but to determine if this would be comparable to wsumsqdiff?

As a side note, I wasn't sure if if should have a separate S<:Number for the w variable instead of use the same T<:Number used for a & b is the input may be an integer or a float separate from variables a and b.

I also wrote (macro) instead of the "at" symbol to avoid pinging anyone with that user name.

squared L2 distance
function sqL2dist{T<:Number}(a::AbstractArray{T}, b::AbstractArray{T})
n = length(a)
length(b) == n || throw(DimensionMismatch("Input dimension mismatch"))
r = 0.0
for i = 1:n
(macro)inbounds r += abs2(a[i] - b[i])
end
return r
end

here is my modified version:

squared L2 distance with weights i.e. sum( w[i] * abs2(a-b) )
function wsqL2dist{T<:Number,S<:Number}(a::AbstractArray{T}, b::AbstractArray{T}, w::AbstractArray{S})
n = length(a)
length(b) == n || throw(DimensionMismatch("Input dimension mismatch"))
length(w) == n || throw(DimensionMismatch("Input dimension mismatch"))
r = 0.0
for i = 1:n
(macro)inbounds r += w[i] * abs2(a[i] - b[i])
end
return r
end

Here is my version that incorporates the inverse (1/w) of the weights:

squared L2 distance with inverse weights i.e. sum( (1/w[i]) * abs2(a-b) )
function winvsqL2dist{T<:Number,S<:Number}(a::AbstractArray{T}, b::AbstractArray{T}, c::AbstractArray{S})
n = length(a)
length(b) == n || throw(DimensionMismatch("Input dimension mismatch"))
length(w) == n || throw(DimensionMismatch("Input dimension mismatch"))
r = 0.0
for i = 1:n
(macro)inbounds r += (1/w[i]) * abs2(a[i] - b[i])
end
return r
end

[timema]

If found that in 0.3 my wsqL2dist function was on par with wsumsqdiff. However I found that if I write:

function sumlog {T<:Number}{a::AbstractArray{T}}
r = 0.0
for i = 1:size(a,1)
(macro)inbounds r += log(a[i])
end
return r
end

this sumlog(x) function take about about 60% of the time that sum(log(x)) does.