Using wrong loop order

[dcelisgarza]

Hi, I'm using the package primarily for its moving averages. I found that pretty much every single case uses the wrong loop order for best performance. The fix is pretty easy, ie just swap the inner and outer loops. For example, this function in movingaverages.jl.

function sma(ta::TimeArray, n::Integer)
  tstamps = timestamp(ta)[n:end]

  vals = zeros(size(values(ta),1) - (n-1), size(values(ta),2))
  for i in 1:size(values(ta),1) - (n-1)
    for j in 1:size(values(ta),2)
      vals[i,j] = nanmean(values(ta)[i:i+(n-1),j])[1]
    end
  end

  cname = Symbol[]
  cols  = colnames(ta)
  for c in 1:length(cols)
    push!(cname, Symbol(string(cols[c], "_sma_", n)))
  end

  TimeArray(tstamps, vals, cname, meta(ta))
end

would simply turn into

function sma(ta::TimeArray, n::Integer)
  tstamps = timestamp(ta)[n:end]

  vals = zeros(size(values(ta),1) - (n-1), size(values(ta),2))
  for j in 1:size(values(ta),2)
    for i in 1:size(values(ta),1) - (n-1)
      vals[i,j] = nanmean(values(ta)[i:i+(n-1),j])[1]
    end
  end

  cname = Symbol[]
  cols  = colnames(ta)
  for c in 1:length(cols)
    push!(cname, Symbol(string(cols[c], "_sma_", n)))
  end

  TimeArray(tstamps, vals, cname, meta(ta))
end