Roger's comment provides a potential solution, but in a comment to me some
time ago he suggested creating the array and then assigning elements to it.
That way you can assign whatever 'zero element' you like, and it solved some
difficult space problems for me.
s =: 3 10 10 NB. array shape
c =: { (<@i.) "0 s NB. catalogue of cell locs
cr =: (15 ?. 300){,c NB. random non zero items
sa=: 1 $. s;(i.#s);1 NB. create empty sparse array
NB. with 1 as 'zero' element
NB. create the array before
NB. assigning values
sa=:(10+i.15) cr} sa NB. assign some 'non zero' elements
0$. sa NB. displays as dense
Fraser
----- Original Message -----
From: "Michael Berry" <[EMAIL PROTECTED]>
To: "'Programming forum'" <[email protected]>
Sent: Sunday, December 17, 2006 1:33 PM
Subject: RE: [Jprogramming] Is there a way I can let J know what
thesparseelement ought to be for the result of a boolean comparison?
Aha! This is the point I did not understand. My large array was sparse.
I
was comparing it (rank 1) to a much smaller dense vector. The smaller
vector
actually is dense, but by coercing it to sparse I get the behavior I was
hoping for. Thank you. (And thank everyone who tries to help a J
novice!)
-Michael
-----Original Message-----
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED] On Behalf Of Roger Hui
Sent: Saturday, December 16, 2006 6:11 PM
To: Programming forum
Subject: Re: [Jprogramming] Is there a way I can let J know what the
sparseelement ought to be for the result of a boolean comparison?
You have drawn incorrect conclusions from your
experiments, or you did not do sufficient experiments.
x=: 4 (1000 [EMAIL PROTECTED] 2e9)} 1 $. 2e9
y=: 4 (1000 [EMAIL PROTECTED] 2e9)} 1 $. 2e9
3 $. x = y
1
3 $. x ~: y
0
3 $. 2 + x
2
3 $. x - 1
_1
3 $. 0 > x - 1
1
3 $. ^ x
1
3 $. ^. x
__
If x and y are sparse and f is atomic, then
(3 $. f y) = f 3$.y
(3 $. x f y) = (3$.x) f 3$.y
that is, the sparse element of f y is f applied
to the sparse element of y; likewise the sparse
element of x f y is (sparse element of x) f
(sparse element of y).
You can use (3;e)$.y to specify that the sparse
element should be e and 8$.y to "remove" the
sparse items. However, both of these are "after
the fact". You should avoid dense f sparse and
sparse f dense if you want to keep everything
sparse.
t=: (1e6 [EMAIL PROTECTED] 100) > $. 1e6{.1
+/t
989986
t
0 │ 1
1 │ 1
2 │ 1
3 │ 1
4 │ 1
...
q=: 8 $. (3;1) $. t
7!:5 ;:'t q'
5243072 320
----- Original Message -----
From: Michael Berry <[EMAIL PROTECTED]>
Date: Saturday, December 16, 2006 1:40 pm
Subject: [Jprogramming] Is there a way I can let J know what the sparse
element ought to be for the result of a boolean comparison?
Dear J experts,
In the course of trying to figure out why I sometimes get "out of
memory"errors when I didn't think I had done anything to cause my
sparse arrays to
become dense, I became curious about how J knows what the right sparse
element will be for the result of a Boolean function where one
argument is a
sparse integer array. By experiment, I see that if I compare my
sparsearray to a scalar argument, the resulting Boolean array has
the appropriate
sparse element, namely the result of comparing the integer array's
sparseelement with the scalar argument:
v1 =.$.100{.3
v2=.$._100{.3
b1=.v1<3
b2=.v2<3
b1
0 | 0
b2
99 | 0
b3=.v1>0
b4=.v2>0
b3
0 | 1
b4
99 | 1
If, on the other hand, I have an array argument, the sparse
element of the
resulting array is always 0:
v5=. ?. 100$10
b1 =. v1>v5
b2 =. v1<v5
3 $. b1
0
3 $. b2
0
+/b1
0
+/b2
94
This explains my out of memory problem. b2 is mostly 1, but its
sparseelement is 0. It is in the nature of the comparisons I am
doing, that I
expect my results to be either mostly 1 or mostly 0 and I know in
advancewhich. I can, of course, rephrase my comparisons so that
the "mostly" case
is always 0, but this sometimes seems unnatural. I don't suppose
there is a
way I could whisper in J's ear to say I'm expecting mostly 1?
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
