Groups


68 of 99+  



julia-users ›
Array comprehension cannot detect type?
8 posts by 4 authors  



rizal zaini 	

11/9/15


Hi all

I want to create matrix containing dot products

julia> X = rand(5,5)
5x5 Array{Float64,2}:
 0.274799   0.564619  0.915595  0.341059  0.57361
 0.0806822  0.293939  0.379279  0.608582  0.652441
 0.264678   0.532751  0.289088  0.457759  0.492777
 0.33856    0.495587  0.499099  0.548237  0.402534
 0.686305   0.168385  0.8322    0.605532  0.518693

julia>  K = [dot(X[:,i], X[:,j]) for i=1:5, j=1:5]
5x5 Array{Any,2}:
 0.737716  0.603229  1.09884  0.865174  0.832958
 0.603229  0.962978  1.16994  0.988988  1.06501
 1.09884   1.16994   2.00739  1.45298   1.54767
 0.865174  0.988988  1.45298  1.36347   1.35304
 0.832958  1.06501   1.54767  1.35304   1.42861

I expect to get an Array{Float64,2}, but what I got is Array{Any,2}. 
How to get Array{Float64,2} using Array comprehension?

Thanks before
Rizal
 


rizal zaini 	

11/9/15


My concern is also on parallel code: 

julia> K = @parallel [dot(X[:,i], X[:,j]) for i=1:5, j=1:5]
5x5 DArray{Any,2,Array{Any,2}}:
1.35134  1.36439  1.28474  1.13181  1.07863
1.36439  1.40077  1.25708  1.22365  1.08332
1.28474  1.25708  1.54494  0.95311  1.15779
1.13181  1.22365  0.95311  1.70348  1.30076
1.07863  1.08332  1.15779  1.30076  1.34577

I expect to get  a DArray{Float64,2,Array{Float64,2}} rather than DArray{Any,2,Array{Any,2}}

Please help. 

Rizal
- show quoted text -
 


Kristoffer Carlsson 	

11/9/15


Don't run the code in global scope or alternatively be explicit about the type you want, for example Float64[....]
- show quoted text -
 


Glen O 	

11/9/15


As far as I'm aware, there are two nice ways to tell the comprehension the type that is going to be returned.

Option 1: Tell it the type of the comprehension:

K = Float64[dot(X[:,i], X[:,j]) for i=1:5, j=1:5]


Option 2: Assert the type of the return of "dot"

K = [dot(X[:,i],X[:,j])::Float64 for i=1:5, j=1:5]

As far as I can tell, they'll do the same thing.
- show quoted text -
 


rizal zaini 	

11/9/15


When I put that code inside a function, I got the same behavior.

For single process, K = Float64[dot(X[:,i], X[:,j]) for i=1:5, j=1:5] does the job.
But for parallel I got error: "ERROR: malformed @parallel loop"

Rizal
- show quoted text -
 


John Gibson 	

11/9/15


I'm no Julia expert, but I'm pretty certain this fall under the category "If you Julia to optimize, put your code in a function."

 gibson@timaeus$ ~/packages/julia-0.4.0/bin/julia
               _
   _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: http://docs.julialang.org
   _ _   _| |_  __ _   |  Type "?help" for help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.4.0 (2015-10-08 06:20 UTC)
 _/ |\__'_|_|_|\__'_|  |  Official http://julialang.org/ release
|__/                   |  x86_64-unknown-linux-gnu

julia> X = rand(2,2)
2x2 Array{Float64,2}:
 0.713809  0.0650767
 0.916874  0.111741 

julia> K = [dot(X[:,i], X[:,j]) for i=1:2, j=1:2]
2x2 Array{Any,2}:
 1.35018   0.148905 
 0.148905  0.0167211

julia> f(X) = [dot(X[:,i], X[:,j]) for i=1:2, j=1:2]
f (generic function with 1 method)

julia> f(X)
2x2 Array{Float64,2}:
 1.35018   0.148905 
 0.148905  0.0167211

John
- show quoted text -
 


Glen O 	

11/9/15


Does the assertion form work any better?

K=[dot(X[:,i],X[:,j])::Float64 for i=1:5, j=1:5]
- show quoted text -
 


rizal zaini 	

11/9/15


It does work. 

Thank you so much Glen. 

Rizal
- show quoted text -