From PDLPorters/pdl#34, better to have separately:
@duffee writes: I'm starting to think that a blog post using Test::PDL to test whether I truely understand how a $matrix * $vector operation works. (Advent calendar warning)
As a starter for an Advent posting, the newly-incorporated Test::PDL, which I'm using to replace all tapprox and nearly all instances of these constructs in the codebase:
approxok all($x == $y)is_deeply [$p->unpdl], [...]ok sum(abs($out-$in)) < 1e-4is $out == $in, 1ok tapprox(...)ok all approx(...)ok all($x == $y)is( int(at(sum($c-$ans))), 0, "conv2d()" )ok +($ans[0] == 50) && ($ans[1] == 1) && ($ans[2] == 2)
... is really good. It checks all types (unless you say not to) and dims are identical, whether NaNs/badvalues/infinities are all the same, and when something fails, you get a message like:
# Failed test 'inplace sorting with bad values'
# at Basic/t/ufunc.t line 56.
# 1/10 values do not match
# got: Double D [10] (PB ) [0 1 2 4 5 6 7 8 BAD BAD]
# expected: Double D [10] (PB ) [0 1 2 4 5 6 7 8 9 BAD]
# First <=5 values differ at:
# [
# [8]
# ]
# Those 'got' values: [BAD]
# Those 'expected' values: [9]
That one was from finding a bug in qsort that had lain hidden, because all(pdl('5 3 BAD') == pdl('5 BAD 2')) returns 1; the BAD does not make all return false because there are any true values.
From PDLPorters/pdl#34, better to have separately:
@duffee writes: I'm starting to think that a blog post using Test::PDL to test whether I truely understand how a $matrix * $vector operation works. (Advent calendar warning)
As a starter for an Advent posting, the newly-incorporated Test::PDL, which I'm using to replace all
tapproxand nearly all instances of these constructs in the codebase:approxok all($x == $y)is_deeply [$p->unpdl], [...]ok sum(abs($out-$in)) < 1e-4is $out == $in, 1ok tapprox(...)ok all approx(...)ok all($x == $y)is( int(at(sum($c-$ans))), 0, "conv2d()" )ok +($ans[0] == 50) && ($ans[1] == 1) && ($ans[2] == 2)... is really good. It checks all types (unless you say not to) and dims are identical, whether
NaNs/badvalues/infinities are all the same, and when something fails, you get a message like:That one was from finding a bug in
qsortthat had lain hidden, becauseall(pdl('5 3 BAD') == pdl('5 BAD 2'))returns 1; theBADdoes not makeallreturn false because there are any true values.