Add precess_cd function #16

TestSubjector · 2017-06-03T19:11:53Z

*TODO.md: Remove 'precess_cd' from list.
*docs/src/ref.md: Add "precess_cd" entry to the manual.
*src/utils.jl: Add precess_cd entry to "utils".
*src/precess_cd.jl: Contains code of precess_cd function.
*test/util-tests.jl: Include tests for precess_cd.

*TODO.md: Remove 'precess_cd' from list. *docs/src/ref.md: Add "precess_cd" entry to the manual. *src/utils.jl: Add precess_cd entry to "utils". *src/precess_cd.jl: Contains code of precess_cd function. *test/util-tests.jl: Include tests for precess_cd.

codecov-io · 2017-06-03T19:26:06Z

Codecov Report

Merging #16 into master will decrease coverage by 0.17%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master      #16      +/-   ##
==========================================
- Coverage   99.71%   99.54%   -0.18%     
==========================================
  Files          61       62       +1     
  Lines        1055     1095      +40     
==========================================
+ Hits         1052     1090      +38     
- Misses          3        5       +2

Impacted Files	Coverage Δ
src/precess_cd.jl	`100% <100%> (ø)`
src/ten.jl	`85.71% <0%> (-14.29%)`	⬇️
src/types.jl	`82.35% <0%> (-5.15%)`	⬇️
src/get_date.jl	`100% <0%> (ø)`	⬆️
src/helio_jd.jl	`100% <0%> (ø)`	⬆️
src/adstring.jl	`100% <0%> (ø)`	⬆️
src/gal_uvw.jl	`100% <0%> (ø)`	⬆️
src/helio_rv.jl	`100% <0%> (ø)`	⬆️
src/precess.jl	`100% <0%> (ø)`	⬆️
src/precess_xyz.jl	`100% <0%> (ø)`	⬆️
... and 6 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 9deab8a...1c65feb. Read the comment docs.

giordano · 2017-06-03T21:53:37Z

src/precess_cd.jl

+function _precess_cd{T<:AbstractFloat}(cd::Matrix{T}, epoch1::T, epoch2::T, crval_old::Vector{T},
+                                       crval_new::Vector{T}, FK4::Bool)
+    crvalold = deg2rad.(crval_old)
+    crvalnew = deg2rad.(crval_new)


Julia sind and cosd functions that allows you to avoid defining crvalold and crvalnew.

giordano · 2017-06-03T22:01:25Z

src/precess_cd.jl

+        c = t*(20043.109 - st*(85.33 + st*0.217) + t*(-42.665 - st*0.217 - t*41.8))/3600
+    end
+    pole_ra = 0
+    pole_dec = 90


pole_ra and pole_dec are type-unstable! Use

pole_ra = zero(T) pole_dec = one(T) * 90

instead ;-)

giordano · 2017-06-03T22:03:33Z

src/precess_cd.jl

+        pole_ra, pole_dec = precess(pole_ra, pole_dec, epoch1, epoch2, FK4=FK4)
+    end
+    sind1, sind2 = sin.([crvalold[2], crvalnew[2]])
+    cosd1, cosd2 = cos.([crvalold[2], crvalnew[2]])


This writing is concise, but you're also allocating useless arrays. You can simply do

sind1 = sind(crval_old[2]) sind2 = ...

(remember the comment above about sind and cosd)

giordano · 2017-06-03T22:06:49Z

src/precess_cd.jl

+    sinra = sin(crvalnew[1] - deg2rad.(pole_ra))
+    cosfi = (cos(c) - sind1*sind2)/(cosd1*cosd2)
+    sinfi = (abs(sin(c))* sinra)/cosd1
+    r = [[cosfi -sinfi]; [sinfi cosfi]]


This can be simplified to

r = [cosfi -sinfi; sinfi cosfi]

which is also much faster and allocates a single array, instead of 3.

giordano · 2017-06-03T22:15:31Z

src/precess_cd.jl

@@ -0,0 +1,78 @@
+# This file is a part of AstroLib.jl. License is MIT "Expat".
+
+function _precess_cd{T<:AbstractFloat}(cd::Matrix{T}, epoch1::T, epoch2::T, crval_old::Vector{T},


For more generality, use AbstractMatrix and AbstractVector in place of Matrix and Vector ;-)

giordano · 2017-06-03T22:15:45Z

src/precess_cd.jl

+### Example ###
+
+```julia
+julia> precess_cd([[20 60]; [45 45]], 1950, 2000, [34, 58], [12, 83])


[[20 60]; [45 45]] can be simplified to [20 60; 45 45]. Also in tests.

giordano · 2017-06-03T22:20:44Z

src/precess_cd.jl

+This function should not be used for values more than 2.5 centuries from the year 1900.
+This function calls [precess](@ref) and [bprecess](@ref).
+"""
+precess_cd{R<:Real}(cd::Matrix{R}, epoch1::Real, epoch2::Real, crval_old::Vector{R},


Like above, for more generality use AbstractMatrix and AbstractVector in place of Matrix and Vector. In addition, here we're not interested in annotating the R type so exploiting new syntax you can simply write

precess_cd(cd::AbstractMatrix{<:Real}, epoch1::Real, epoch2::Real, crval_old::AbstractVector{<:Real}, crval_new::AbstractVector{<:Real}, FK4::Bool=false) =

This couldn't be done in previous Julia <= 0.5. Have a look at the third point of "New type system capabilities" at https://github.com/JuliaLang/julia/blob/release-0.6/NEWS.md#new-language-features

giordano · 2017-06-03T22:29:37Z

src/precess_cd.jl

+    cosfi = (cos(c) - sind1*sind2)/(cosd1*cosd2)
+    sinfi = (abs(sin(c))* sinra)/cosd1
+    r = [[cosfi -sinfi]; [sinfi cosfi]]
+    cd = cd * r


Pay attention here! You're mutating the input cd array. This may a good idea, but if you do so, you should rename the function to precess_cd! (and document that the input array is mutated), because the exclamation mark appended to function name is used to denote a mutating function. Furthermore, with improved loop fusion this could be better written as cd .= cd * r

On a second thought, I think that's probably better not to mutate the input array. Thus, I suggest to delete the assignment cd = ... and simply return cd * r (but also look at the order of operands, I'm not sure it's correct).

For the record, this assignment does not mutate the input array. It first creates a new array cd * r then assigns the value to the identifier cd. The identifier cd no longer points to the input array, which is unmodified.

But yeah, simply return cd * r would be good here.

Uhm, you're right! I was doing tests and noted that the input cd matrix was changing, but probably I had replaced the line with cd .= cd * r, which does mutate.

In precess_cd.pro, the order of operands is r # cd (r * cd). However, from the IDL documentation, the matrix multiplication in IDL is done by multiplying the columns of the first array by the rows of the second array. So I reversed the the operands to make it a normal matrix multiplication.

giordano · 2017-06-03T22:39:07Z

test/utils-tests.jl

@@ -441,6 +441,22 @@ let
    @test dec2 ≈ [-56.87186126487889]
 end

+# Test precess_cd
+@testset "precess_cd" begin
+    @test precess_cd([[30 60]; [60 90]], 1950, 2000, [13, 8], [43, 23]) ≈


Did you compare the results of these tests with IDL AstroLib?

Nope, haven't been able to run IDL on my distro till now.

Check out http://gnudatalanguage.sourceforge.net/ then, most AstroLib procedures work in GDL ;-)

TestSubjector · 2017-06-05T01:01:34Z

src/precess_cd.jl

+    sinra = sind(crval_new[1] - pole_ra)
+    cosfi = (cos(c) - sind1*sind2)/(cosd1*cosd2)
+    sinfi = (abs(sin(c))* sinra)/cosd1
+    r = [cosfi sinfi; -sinfi cosfi]


I had transposed this r = [cosfi -sinfi; sinfi cosfi] because I initially wanted to keep operands sequence as r * cd, but then I changed my mind about the sequence and forgot to rectify the transposing. Corrected.

giordano · 2017-06-05T07:53:36Z

src/precess_cd.jl

+
+### Arguments ###
+
+* `cd`: 2 x 2 coordinate description matrix in degrees or radians


It seems that the matrix has to be in degrees. I think that the original description in inaccurate, they also say the matrix is altered to match the units of input matrix, but they don't do so.

giordano · 2017-06-05T08:20:46Z

src/precess_cd.jl

+    else
+        pole_ra, pole_dec = precess(pole_ra, pole_dec, epoch1, epoch2, FK4=FK4)
+    end
+    sind1, sind2 = sind.([crval_old[2], crval_new[2]])


Don't use the arrays here ;-) If you still want to have two calculation in one line you can do

sind1, sind2 = sind(crval_old[2]), sind(crval_new[2]) ...

but I don't think you can gain much with respect to

sind1 = sind(crval_old[2]) sind2 = sind(crval_new[2]) ...

That's up to you

giordano · 2017-06-05T08:32:39Z

src/precess_cd.jl

+
+function _precess_cd{T<:AbstractFloat}(cd::AbstractMatrix{T}, epoch1::T, epoch2::T, crval_old::AbstractVector{T},
+                                       crval_new::AbstractVector{T}, FK4::Bool)
+    t = deg2rad((epoch2 - epoch1)*0.001)


I think that you shouldn't call deg2rad here, t isn't simply a global multiplicative factor in definition of c, which is rather a third-order polynomial in t.

giordano · 2017-06-05T08:33:36Z

src/precess_cd.jl

+
+    if FK4
+        st = (epoch1 - 1900)*0.001
+        c = t*(20046.85 - st*(85.33 + st*0.37) + t*(-42.67 - st*0.37 - t*41.8))/3600


Remove division by 3600 and wrap the whole right-hand-side with sec2rad (this is an AstroLib.jl's function).

giordano · 2017-06-05T08:35:11Z

src/precess_cd.jl

+        c = t*(20043.109 - st*(85.33 + st*0.217) + t*(-42.665 - st*0.217 - t*41.8))/3600
+    end
+    pole_ra = zero(T)
+    pole_dec = one(T)*90


I told you to use one(T) * 90, but T(90) is actually better. Sorry for the wrong suggestion before.

giordano · 2017-06-05T08:35:45Z

src/precess_cd.jl

+        c = t*(20046.85 - st*(85.33 + st*0.37) + t*(-42.67 - st*0.37 - t*41.8))/3600
+    else
+        st = (epoch1 - 2000)*0.001
+        c = t*(20043.109 - st*(85.33 + st*0.217) + t*(-42.665 - st*0.217 - t*41.8))/3600


Same as at line 9.

Test cases values found to be in an acceptable range with those received from precess_cd.pro (using GDL - GNU Data Language [Version 0.9.4])

giordano · 2017-06-06T22:12:41Z

src/precess_cd.jl

+        c = sec2rad(t*(20043.109 - st*(85.33 + st*0.217) + t*(-42.665 - st*0.217 - t*41.8)))
+    end
+    pole_ra = zero(T)
+    pole_dec = one(90)


T(90), not one(90) :-) (you have to adjust the example and tests)

Whoops! Fixed now

* Fixed silly mistake * Updated example and tests

kbarbary · 2017-06-07T20:08:29Z

test/utils-tests.jl

+         5107.504395433958 6887.55936949333   ]
+    @test precess_cd([30.0 28.967; 60.45 90.65], 2000, 1975, [13, 10.658], [35.54, 67], true) ≈
+        [64.78429186351575 62.637156996728194 ;
+         130.49379143419722 195.9699513801844 ]


Were these values generated by running the Julia code or by the IDL version? I think we're really going to want to have tests against the IDL version output. But at the very least, there should be a comment here explaining where the "truth" values come from.

I compared some of the results with IDL AstroLib. There's a difference on the leas significant digits, but it happens the same with bprecess.

I don't think it's a good idea to always compare with results from IDL AstroLib, there are often 32- and 64-bit floating point numbers mixed together, so that they're often numerically unreliable if one expects full 64-bit accuracy

One could then just decrease the required precision for isapprox, right? I think a test against an independent implementation is more valuable than a test of this implementation against itself.

Or we could have both tests, or maybe just put the IDL results in a comment.

One could then just decrease the required precision for isapprox, right?

Yes, and for the specific case probably this isn't even needed, because rtol defaults to sqrt(eps(T)) (pretty high, IMHO) and atol to 0, I think the results will be well consistent within 1 part in ~10^8.

I think a test against an independent implementation is more valuable than a test of this implementation against itself.

If you don't know the "truth", testing an implementation against itself doesn't ensure the code is doing the Right Thing, but at least checks that nothing will be broken by future changes.

Or we could have both tests, or maybe just put the IDL results in a comment.

Commenting the IDL results seems doable. I could work on it during and after the missing astronomical routines are done.

TestSubjector added 2 commits June 4, 2017 00:30

Add precess_cd function

90f8767

*TODO.md: Remove 'precess_cd' from list. *docs/src/ref.md: Add "precess_cd" entry to the manual. *src/utils.jl: Add precess_cd entry to "utils". *src/precess_cd.jl: Contains code of precess_cd function. *test/util-tests.jl: Include tests for precess_cd.

Quick file name fix

138adda

Fix test for FK4 frame

a095ad3

giordano requested changes Jun 3, 2017

View reviewed changes

Improvements and Correcting matrix r

221b889

TestSubjector commented Jun 5, 2017

View reviewed changes

giordano reviewed Jun 5, 2017

View reviewed changes

Final fixes

bfae37c

Test cases values found to be in an acceptable range with those received from precess_cd.pro (using GDL - GNU Data Language [Version 0.9.4])

giordano reviewed Jun 7, 2017

View reviewed changes

TestSubjector added 2 commits June 7, 2017 16:51

Replaced one(90) with T(90)

520dd90

* Fixed silly mistake * Updated example and tests

Add reference to sec2rad

2e611f9

kbarbary reviewed Jun 7, 2017

View reviewed changes

Comment information about test values

1c65feb

giordano merged commit c85a71a into JuliaAstro:master Jun 9, 2017

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add precess_cd function #16

Add precess_cd function #16

TestSubjector commented Jun 3, 2017

codecov-io commented Jun 3, 2017 •

edited

Loading

giordano Jun 3, 2017

giordano Jun 3, 2017

giordano Jun 3, 2017

giordano Jun 3, 2017

giordano Jun 3, 2017

giordano Jun 3, 2017

giordano Jun 3, 2017

giordano Jun 3, 2017

giordano Jun 4, 2017

kbarbary Jun 4, 2017

giordano Jun 4, 2017

TestSubjector Jun 5, 2017

giordano Jun 3, 2017

TestSubjector Jun 5, 2017

giordano Jun 5, 2017

TestSubjector Jun 5, 2017

giordano Jun 5, 2017

giordano Jun 5, 2017 •

edited

Loading

giordano Jun 5, 2017 •

edited

Loading

giordano Jun 5, 2017

giordano Jun 5, 2017

giordano Jun 5, 2017

giordano Jun 6, 2017

TestSubjector Jun 7, 2017 •

edited

Loading

kbarbary Jun 7, 2017

giordano Jun 7, 2017

kbarbary Jun 7, 2017

giordano Jun 7, 2017

TestSubjector Jun 8, 2017

		@@ -0,0 +1,78 @@
		# This file is a part of AstroLib.jl. License is MIT "Expat".

		function _precess_cd{T<:AbstractFloat}(cd::Matrix{T}, epoch1::T, epoch2::T, crval_old::Vector{T},


		### Arguments ###

		* `cd`: 2 x 2 coordinate description matrix in degrees or radians

Add precess_cd function #16

Add precess_cd function #16

Conversation

TestSubjector commented Jun 3, 2017

codecov-io commented Jun 3, 2017 • edited Loading

Codecov Report

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

giordano Jun 5, 2017 • edited Loading

Choose a reason for hiding this comment

giordano Jun 5, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

TestSubjector Jun 7, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov-io commented Jun 3, 2017 •

edited

Loading

giordano Jun 5, 2017 •

edited

Loading

giordano Jun 5, 2017 •

edited

Loading

TestSubjector Jun 7, 2017 •

edited

Loading