VEP of advanced products

[bkhelifi]

Dear all,

Before starting the discussion into the Semantics WG, here is the VEP for the advanced products samples, pdf and region.

Any comment is welcome. The deadline is the 25th of February.

VEP-advancedproducts.txt

[BaptisteCecconi]

Hi @bkhelifi!
I had a look to the proposed advanced product. The first one is samples. I understand the background of the name and the definition but it will conflict with an other use of the term sample, which is used in the context of physical sample analysis from astronomical objects (meteors, comets, asteroids...).

[iannevans]

Hi Baptiste, Well, I’m really confused why the term “samples” being proposed here, since it is *not* the term we proposed in the draft HEIG ObsCore extension document. There we used the term “draws”, which is the more appropriate term in my opinion for a couple of reasons. (1) The term “samples" is very general and is widely used in astronomy for a variety of different purposes (and as you indicate, it’s very close to “sample” and can easily be confused). (2) There is a subtle difference between the widely used meanings of the term “samples” used in statistical analyses and the term “draws”, although they are often used interchangeably. “Samples” are the individual components of a statistical sample selected from a larger population, and the sample is typically used as representative of a population. This term is commonly used in frequentist statistical analyses. “Draws” are very similar, but can be drawn either from a population or from a probability distribution (such as the posterior probability distribution used in Bayesian statistics). This term is commonly used in Bayesian statistical analyses, *but is also applicable to frequentist analyses* (in the former case is sampling parameters of the distribution whereas for the latter case one is sampling data points from the observed population). In both cases, the process of selecting samples or draws is called “sampling” and the algorithm used is term a “sampler” (just to confuse things :-). Because of this, I continue to recommend the use of the term “draws”. I note that the existing datasets that require this definition are Bayesian posterior distributions where “samples” isn’t really an appropriate choice. The last paragraph of the draft Rationale for the term samples that was distributed previously is not consistent with the widely used meanings of the terms “samples” and “draws”. Thanks, —Ian

On Feb 16, 2026, at 08:24, Baptiste Cecconi ***@***.***> wrote: BaptisteCecconi left a comment (ivoa/HighEnergyObsCoreExt#38) <https://www.google.com/url?q=https://github.com/ivoa/HighEnergyObsCoreExt/issues/38%23issuecomment-3908488045&source=gmail-imap&ust=1771853043000000&usg=AOvVaw2RIayAsgFoxINS-5jcCEnS> Hi @bkhelifi <https://www.google.com/url?q=https://github.com/bkhelifi&source=gmail-imap&ust=1771853043000000&usg=AOvVaw1mO_bcI9ABh35JXzw2cN7g>! I had a look to the proposed advanced product. The first one is samples. I understand the background of the name and the definition but it will conflict with an other use of the term sample, which is used in the context of physical sample analysis from astronomical objects (meteors, comets, asteroids...). — Reply to this email directly, view it on GitHub <https://www.google.com/url?q=https://github.com/ivoa/HighEnergyObsCoreExt/issues/38%23issuecomment-3908488045&source=gmail-imap&ust=1771853043000000&usg=AOvVaw2RIayAsgFoxINS-5jcCEnS>, or unsubscribe <https://www.google.com/url?q=https://github.com/notifications/unsubscribe-auth/ACCWIHA66IO437QNZLPFGMT4MHAHFAVCNFSM6AAAAACVIY3IJ2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTSMBYGQ4DQMBUGU&source=gmail-imap&ust=1771853043000000&usg=AOvVaw1ow-5KO50N7hD8cH08LYej>. You are receiving this because you are subscribed to this thread.

— Dr. Ian Evans Astrophysicist Chandra X-ray Center Center for Astrophysics | Harvard & Smithsonian Office: (617) 496 7846 | Cell: (617) 699 5152 60 Garden Street | MS 81 | Cambridge, MA 02138   <http://cfa.harvard.edu/>cfa.harvard.edu <http://cfa.harvard.edu/> | Facebook <http://cfa.harvard.edu/facebook> | Twitter <http://cfa.harvard.edu/twitter> | YouTube <http://cfa.harvard.edu/youtube> | Newsletter <http://cfa.harvard.edu/newsletter>

[loumir]

Thanks for the proposal for terms to designate advanced products types.
I think we need examples of the various advanced data products, as real files ,with explanations of the content, to stabilize this vocabulary.
Other energy domains as in the radio domain , have similar needs to expose complementary data sets in order to characterize the statistical behaviors of the data.

This was an idea I suggested at a HEIG Meeting some time ago.
having a table with a set of file examples of each category would help .

I suggest we have it in a subdirectory in this git repo.
if the files are too big , we can provide a link to it .

[bkhelifi]

Hi Ian,

One have 2 aspects in this VEP:

• the description of the concept
• the choosen workind of the concept

First, is the current definition fine with you?

Concerning the wording, this is hard, I admit that.

In both cases, the process of selecting samples or draws is called “sampling” and the algorithm used is term a “sampler” (just to confuse things :-).
This is one main reason I made the proposal.

The second is that the term 'draws' is too Bayesian in connotation, at least from what I have discovered in the litterature. It is a connotation only, because I have seen also draws used for Frequentist stats (as you have mentionned. But, from my personal small study, samples seems to be more generalist. I have even asked to some AI bots to get an other idea, even if one should take a huge care of them!
But I am not a statistician and I would be great to have some advice.

Let's see what thinks the rest of the community... about the 2 words samples or draws.

PS: do not forget that we will have an associated UCD to this product, that will be stat.likelihood or stat.probability. This will leviate any ambiguity with other vocabularies in other domain. But in statistics, samples is a key semantic word.

[msdemlei]

On Mon, Feb 16, 2026 at 10:22:54AM -0800, Bruno Khélifi wrote: bkhelifi left a comment (ivoa/HighEnergyObsCoreExt#38) Let's see what thinks the rest of the community... about the 2 words `samples` or `draws`.

I, for one, am on the side of "draws". One shouldn't sweat identifiers and labels *too* much (the concept is what the definition says), but making them so that we don't set up traps is wise. I could well see "samples" looking like a desirable choice for some data set from the laboratory, and so I prefer we don't hog it.

[iannevans]

Hi Ian,

One have 2 aspects in this VEP:

• the description of the concept
• the choosen workind of the concept

First, is the current definition fine with you?

No, it's not.

We very explicitly defined the vocabulary terms in section 5.1 of the draft HEIG ObsCore Extension note, and ensured that they matched the dataproduct_type definitions in section 3.1. We agree to these definitions many months ago, and a good chunk of the note is based on this (they're used and referenced in many sections and tables). The definitions of the terms in the VEPs (which I assume is what is listed under "Description") need to match the vocabulary definitions in the note. Otherwise we need to go back and revise the note for consistency.

With regard to the Rationale, I need to review and think about the wording in more detail but we would not often use the median of the posterior probability as the best estimate of a parameter value (more typically would use the mode). But in any case I don't think the rationale as stated really explains why one wants the draws as an identified type of data product, i.e., what is unique about a draws data products that you couldn't get from some other source (e.g.) a pdf data product or mean and variance values etc. I think the rationale's need to explain why the data product is important and why users will want to identify and access this kind of data product.

Concerning the wording, this is hard, I admit that.

In both cases, the process of selecting samples or draws is called “sampling” and the algorithm used is term a “sampler” (just to confuse things :-).
This is one main reason I made the proposal.

The second is that the term 'draws' is too Bayesian in connotation, at least from what I have discovered in the litterature. It is a connotation only, because I have seen also draws used for Frequentist stats (as you have mentionned. But, from my personal small study, samples seems to be more generalist. I have even asked to some AI bots to get an other idea, even if one should take a huge care of them! But I am not a statistician and I would be great to have some advice.

Sorry, but I don't agree. "Draws" is used for both frequentist and Bayesian analyses, although it's probably seen more for Bayesian analysis because many frequentist analyses do use "samples". You will rarely find "samples" used for Bayesian analyses (although I have seen it occasionally) because "samples" from a probability distribution don't really make sense because it's a continuous variable, whereas in frequentist analysis you are conceptually dealing with a very large (infinite) population of experimental results, which is really a discrete quantity.

And "samples" is really a concept that is used too generically in astrophysics to use for this limited case (kinda like "background").

Let's see what thinks the rest of the community... about the 2 words samples or draws.

PS: do not forget that we will have an associated UCD to this product, that will be stat.likelihood or stat.probability. This will leviate any ambiguity with other vocabularies in other domain. But in statistics, samples is a key semantic word.

[iannevans]

Thanks for the proposal for terms to designate advanced products types. I think we need examples of the various advanced data products, as real files ,with explanations of the content, to stabilize this vocabulary. Other energy domains as in the radio domain , have similar needs to expose complementary data sets in order to characterize the statistical behaviors of the data.

This was an idea I suggested at a HEIG Meeting some time ago. having a table with a set of file examples of each category would help .

I suggest we have it in a subdirectory in this git repo. if the files are too big , we can provide a link to it .

I think it's possible to provide some example data products, but I would not place too strong a weight on those data products when defining the vocabulary because the advanced data products are all very generic, multi-waveband concepts. Relying on how they are used for Chandra X-ray data only for example will almost certainly artificially and inappropriately constrain to vocabulary. I think it is more important here to be thinking broadly based on experience across a wide range of wavebands and technologies.