Oh, didn't catch that you wanted row sums, but yeah, same basic idea.
On Tue, Jul 28, 2015 at 12:03 PM, Tom Breloff <[email protected]> wrote:
> I would just write the loop explicitly:
>
>
> julia> function meanpositives(mat::Matrix)
> nc = size(mat,2)
> sums = zeros(nc)
> counts = zeros(nc)
> for c in 1:nc
> for r in 1:size(mat,1)
> v = mat[r,c]
> if v > 0
> sums[c] += v
> counts[c] += 1.0
> end
> end
> end
> Float64[(counts[i]>0 ? sums[i]/counts[i] : 0.0) for i in 1:nc]
> end
> meanpositives (generic function with 1 method)
>
> julia> x
> 3x3 Array{Float64,2}:
> 10.0 5.0 0.0
> 5.0 0.0 0.0
> 0.0 0.0 0.0
>
> julia> meanpositives(x)
> 3-element Array{Float64,1}:
> 7.5
> 5.0
> 0.0
>
>
> On Tue, Jul 28, 2015 at 12:01 PM, Stefan Karpinski <[email protected]>
> wrote:
>
>> You mean quicker as in faster or as in easier? Using indexing is probably
>> easiest but not fastest:
>>
>> julia> M = randn(5,5)
>> 5x5 Array{Float64,2}:
>> -1.51134 1.2331 -0.186083 0.412282 -1.13114
>> -0.0411132 0.124684 0.377426 0.622427 -0.278162
>> 0.788182 -0.834092 1.09506 1.45196 -0.620717
>> 1.45856 -0.889277 -0.768049 -0.95953 -0.340098
>> 1.69666 0.695566 0.706905 0.63089 0.958474
>>
>> julia> mean(M[M .> 0])
>> 0.8751552351448673
>>
>>
>> If you want the fastest, loop through the matrix and keep totals and
>> counts:
>>
>> julia> function meanpos(a)
>> t = zero(eltype(a))
>> c = 0
>> for x in a
>> x > 0 || continue
>> t += x
>> c += 1
>> end
>> return t/c
>> end
>> meanpos (generic function with 1 method)
>>
>> julia> meanpos(M)
>> 0.8751552351448673
>>
>>
>> This should be fairly efficient for any non-sparse collection.
>>
>>
>> On Tue, Jul 28, 2015 at 11:50 AM, Jude <[email protected]> wrote:
>>
>>> Hi!
>>>
>>> Probably a quite basic question but was wondering if someone knows a quick
>>> way to get the average of theh positive elements of a matrix, eg, If I have
>>> an m*n matrix and want to get the average for each m so have m*1 but only
>>> want to work with non-zero elements and ignore the zero elements and
>>> negative elements
>>>
>>> I guess I could using indexing and pull out the indexes corresponding to
>>> the positive elements but just wondering if anyone knew quicker ways!
>>>
>>> Thanks!
>>>
>>
>>
>
