Allowing using `NamedTuple` as `initial_params` #632

torfjelde · 2024-07-15T09:00:07Z

Quick and dirty implementation, but does the trick.

I probably shouldn't spend much time on this (have other PRs that I need to complete), but figured I'd just add this here in case someone else wants to complete it (maybe @sunxd3 or @mhauru has time) :)

Basically, it's just missing:

Testing

sunxd3 · 2024-07-15T09:08:04Z

Looks reasonable, maybe we can just move update_values!! to out into utils.jl.
Regardless, I can write some tests for this.

Can I get your permission to just commit to this PR? @torfjelde

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

torfjelde · 2024-07-15T09:11:34Z

Looks reasonable, maybe we can just move update_values!! to out into utils.jl.

Indeed:)

Can I get your permission to just commit to this PR? @torfjelde

Yep, go for it!

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

coveralls · 2024-07-15T16:46:23Z

Pull Request Test Coverage Report for Build 10038614257

Details

20 of 24 (83.33%) changed or added relevant lines in 3 files are covered.
13 unchanged lines in 2 files lost coverage.
Overall coverage decreased (-0.05%) to 81.36%

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/sampler.jl	16	17	94.12%
src/varinfo.jl	0	3	0.0%

Files with Coverage Reduction	New Missed Lines	%
src/model.jl	1	87.0%
src/threadsafe.jl	12	45.19%

Totals
Change from base Build 10038508405:	-0.05%
Covered Lines:	2811
Relevant Lines:	3455

💛 - Coveralls

sunxd3

@torfjelde I changed several lines of your code, could you give a quick look? The rationales are in the code review comments.

sunxd3 · 2024-07-16T07:01:16Z

src/sampler.jl

-    vi::AbstractVarInfo, initial_params, spl::Sampler, model::Model
+function set_values!!(
+    varinfo::AbstractVarInfo,
+    initial_params::AbstractVector{<:Union{Real,Missing}},


I changed the array element type from Real to Union{Real, Missing} because we allow initialization vector with missing in it.

Really? In what scenario would this be used?

DynamicPPL.jl/test/sampler.jl

Line 135 in dfdc155

chain = sample(model, sampler, 1; initial_params=[missing, -1], progress=false)

Hmm, not sure how I feel about that. I guess it means to sample that parameter from the prior? But that's so simple to do by hand anyways these days.

I suppose so, although it made sense to me because the initialization vector need to be the same dimension as the model, so it makes sense for someone to say "I don't care about these" by setting them to missing

Sure, but you can also just do

rand(Vector, model)

and alter those values 😕

Basically, it's a question of whether we want to maintain this or just leave it up to the user when it is a simple one-liner.

Got your argument, 👍
But in this case it's not really a one-liner, as user also need to set individual values.
For models with small dimensions, current syntax can still be useful. I am for keeping this option.

sunxd3 · 2024-07-16T07:02:27Z

src/sampler.jl

+end
+
+# if initialize with scalar, convert to vector
+function set_values!!(varinfo::AbstractVarInfo, initial_params::Real, spl::AbstractSampler)


The reason of adding this method is that, before this PR, initial_params can be a scalar.

Real is not concrete, but probably okay given this likely is not on the critical path.

Was not aware we allowed this either 🤷 Are we testing this?

Yeah

DynamicPPL.jl/test/sampler.jl

Line 87 in dfdc155

chain = sample(model, sampler, 1; initial_params=0.2, progress=false)

Honestly, I say we drop this. Seems unnecessarily complicated when it's just a difference between 1 and [1].

I see @torfjelde's point of simplicity of interface, but it is a breaking change, which makes me wonder if the small effort of keeping it supported is worth it. If we do change it, maybe bundle it with some other breaking changes?

Though it's technically breaking, I highly doubt this piece of code exists anywhere else, i.e. it would be a matter of bumping compat bounds 🤔

sunxd3 · 2024-07-16T07:06:07Z

src/sampler.jl

+function set_values!!(
+    varinfo::AbstractVarInfo, initial_params::NamedTuple, spl::AbstractSampler
+)
+    initial_params = NamedTuple(k => v for (k, v) in pairs(initial_params) if v !== missing)


Not sure this is good, but changed to support a more uniform interface: allowing NamedTuple with missing values.

Same as in last comment: when are we initializing using missing?

see reply above

mhauru

Mostly looks great, I had a few questions I raised in local comments.

src/sampler.jl

mhauru · 2024-07-19T10:51:15Z

src/sampler.jl

+end
+
+# if initialize with scalar, convert to vector
+function set_values!!(varinfo::AbstractVarInfo, initial_params::Real, spl::AbstractSampler)


I see @torfjelde's point of simplicity of interface, but it is a breaking change, which makes me wonder if the small effort of keeping it supported is worth it. If we do change it, maybe bundle it with some other breaking changes?

src/sampler.jl

mhauru · 2024-07-19T10:56:23Z

src/sampler.jl

+end
+
+function set_values!!(
+    varinfo::AbstractVarInfo, initial_params::NamedTuple, spl::AbstractSampler


Are we happy with spl not doing anything in this function? I get that we need to leave it there to keep a uniform interface, but could this end up being called with varinfo that has not been sampled with spl, which would then produce unexpected results?

mhauru · 2024-07-19T10:57:18Z

src/varinfo.jl

@@ -892,6 +892,12 @@ Base.keys(vi::TypedVarInfo{<:NamedTuple{()}}) = VarName[]
    return expr
 end

+# FIXME(torfjelde): Don't use `_getvns`.


Is this still relevant? Don't know why _getvns should be shunned.

It's just unnecessary; Base.keys should just be it + IIRC _getvns is overly complicated.

torfjelde · 2024-07-22T08:57:14Z

LGTM:) Unfortunately I seems I can't approve because I'm the creator of the PR 🙃

sunxd3 · 2024-07-22T09:14:21Z

wait till #626 to merge, then I'll fix the versioning and potential errors

sunxd3 · 2024-07-22T10:05:38Z

Do we need to make this breaking?
If not, we can combine the release with #626

Initial work on NamedTuple as initial_params

52a9b76

torfjelde mentioned this pull request Jul 15, 2024

Allow for a named tuple for initial_params TuringLang/Turing.jl#2286

Closed

torfjelde marked this pull request as draft July 15, 2024 09:00

Apply suggestions from code review

94c2ef7

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

sunxd3 and others added 4 commits July 15, 2024 15:54

add some tests

5b5f4dd

Update src/sampler.jl

4693fa2

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

fix type error for inits with nothing

bab789e

Update src/sampler.jl

c3ce3fb

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

sunxd3 reviewed Jul 16, 2024

View reviewed changes

Merge branch 'master' into torfjelde/namedtuple-initial-params

aab6606

sunxd3 requested review from yebai and mhauru July 19, 2024 10:13

sunxd3 marked this pull request as ready for review July 19, 2024 10:55

mhauru reviewed Jul 19, 2024

View reviewed changes

sunxd3 changed the title ~~[DRAFT] Proof-of-concept for initial_params with NamedTuple~~ Allowing using NamedTuple as initial_params Jul 22, 2024

sunxd3 added 4 commits July 22, 2024 08:47

use better variable names

22942d4

remove init with scalar

a3969d8

move update_values!! out of TestUtils

1b2ce47

fix error

14bb2cf

Merge branch 'master' into torfjelde/namedtuple-initial-params

8b79ac0

yebai enabled auto-merge July 23, 2024 12:25

yebai added this pull request to the merge queue Jul 23, 2024

Merged via the queue into master with commit 142f753 Jul 23, 2024
11 of 12 checks passed

yebai deleted the torfjelde/namedtuple-initial-params branch July 23, 2024 14:50

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Allowing using `NamedTuple` as `initial_params` #632

Allowing using `NamedTuple` as `initial_params` #632

torfjelde commented Jul 15, 2024 •

edited by sunxd3

Loading

sunxd3 commented Jul 15, 2024

torfjelde commented Jul 15, 2024

coveralls commented Jul 15, 2024 •

edited by github-actions bot

Loading

sunxd3 left a comment

sunxd3 Jul 16, 2024

torfjelde Jul 16, 2024

sunxd3 Jul 16, 2024

torfjelde Jul 16, 2024

sunxd3 Jul 16, 2024

torfjelde Jul 18, 2024

sunxd3 Jul 22, 2024

sunxd3 Jul 16, 2024

torfjelde Jul 16, 2024

sunxd3 Jul 16, 2024

torfjelde Jul 18, 2024

mhauru Jul 19, 2024

torfjelde Jul 19, 2024

sunxd3 Jul 16, 2024

torfjelde Jul 16, 2024

sunxd3 Jul 16, 2024

mhauru left a comment

mhauru Jul 19, 2024

mhauru Jul 19, 2024

mhauru Jul 19, 2024

torfjelde Jul 19, 2024

torfjelde commented Jul 22, 2024

sunxd3 commented Jul 22, 2024

sunxd3 commented Jul 22, 2024 •

edited

Loading

Allowing using NamedTuple as initial_params #632

Allowing using NamedTuple as initial_params #632

Conversation

torfjelde commented Jul 15, 2024 • edited by sunxd3 Loading

sunxd3 commented Jul 15, 2024

torfjelde commented Jul 15, 2024

coveralls commented Jul 15, 2024 • edited by github-actions bot Loading

Pull Request Test Coverage Report for Build 10038614257

Details

💛 - Coveralls

sunxd3 left a 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

Choose a reason for hiding this comment

mhauru left a 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

torfjelde commented Jul 22, 2024

sunxd3 commented Jul 22, 2024

sunxd3 commented Jul 22, 2024 • edited Loading

Allowing using `NamedTuple` as `initial_params` #632

Allowing using `NamedTuple` as `initial_params` #632

torfjelde commented Jul 15, 2024 •

edited by sunxd3

Loading

coveralls commented Jul 15, 2024 •

edited by github-actions bot

Loading

sunxd3 commented Jul 22, 2024 •

edited

Loading