Is it possible to use Arrayfun across rows

13 次查看(过去 30 天)
Hi,
I currently have a FOR LOOP which works its way through a table with almost 20 million records. It is as expected pretty slow, I want to look into alternatives and I wondered if there is a way to use for arrayfun - or another MATLAB function - across rows which will work with high performance. The example below captures the issue of working across rows:
A = table([1;1;1;2;2;2;],[1;2;3;4;5;6]);
A.Var3 = zeros(height(A),1)
A.Var3(1) = A.Var1(1)
for i = 2:height(A)
if A.Var1(i) == A.Var1(i-1)
A.Var3(i) = A.Var2(i) .* A.Var2(i-1);
else A.Var3(i) = A.Var2(i);
end
end
Any suggestions will be appreciated.
Kind regards,
William
  11 个评论
Rik
Rik 2020-10-6
The longer the runs are, the more efficient calculating the runs will be. So if you have long stretches of true and/or long stretches of false it might be worth looking into. I think the first branch can also be vectorized (e.g. with cumprod), although I haven't tried yet.
William Ambrose
William Ambrose 2020-10-6
You are right with respect to the true/false, I will definitly use
i = [false; A.Var1(1:end-1) == A.Var1(2:end)]
to get the logical array, I suspect it will make a big impact.

请先登录,再进行评论。

回答(1 个)

Mohammad Sami
Mohammad Sami 2020-10-6
Something like this will work.
i = [false; A.Var1(1:end-1) == A.Var1(2:end)];
j = find(i);
A.Var3(i) = A.Var2(j) .* A.Var2(j-1);
A.Var3(~i) = A.Var2(~i);
  5 个评论
William Ambrose
William Ambrose 2020-10-8
Hi Mohammed,
I like the use of grouptransform but unfortunately the use of @cumprod as the function isn't correct. What I need is an iterative cascade through the rows where the result in the previous row - in another column - is input into the calculation. Almost like a FOR LOOP solution for the Fibonnaci sequence.
To be sure I will include the corrected version of the code, with some of your suggestions embedded in, and expected outcome:
B = table([1;1;1;1;1;2;2;2;3],[1;2;3;4;5;6;7;8;500]);
B.Var3 = zeros(height(B),1);
i = [false; B.Var1(1:end-1) == B.Var1(2:end)];
j = find(~i);
B.Var3(j) = B.Var2(j);
tic;
for n = 1:height(B)
if i(n) == 1
B.Var3(n) = B.Var3(n-1) .* B.Var2(n);
end
end ; toc
and the result
B =
9×3 table
Var1 Var2 Var3
____ ____ ____
1 1 1
1 2 2
1 3 6
1 4 24
1 5 120
2 6 6
2 7 42
2 8 336
3 500 500
I have appreciate all your suggestions so far so if you have input which could solve the above but without the FOR LOOP I would be very greatful for any input.
Mohammad Sami
Mohammad Sami 2020-10-8
Hi William,
For the updated problem as stated, grouptransform with cumprod will work just as well.
My testing shows the result is identical to the expected result.
A =
9×3 table
Var1 Var2 fun_Var2
____ ____ ________
1 1 1
1 2 2
1 3 6
1 4 24
1 5 120
2 6 6
2 7 42
2 8 336
3 500 500
Ofcourse if the formula changes, for loop may be more generalizable.

请先登录,再进行评论。

类别

Help CenterFile Exchange 中查找有关 Matrices and Arrays 的更多信息

产品


版本

R2019b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by