Project

General

Profile

Actions

CdmVersionTwoDiscussion » History » Revision 62

« Previous | Revision 62/77 (diff) | Next »
Andreas Müller, 03/31/2009 06:09 PM


CDM v2.0 Discussion


This is a site to discuss possible changes to the CDM v1.4 to go into CDM v2.0

See also Component C5.80 - Review of CDM v.1 and model for descriptive data in CDM v.2


Taxonomic Data

=== Separation Of Taxonomic Concept And Taxonomic View

There is a need to separate taxonomic concept information from taxonomic view information resulting in the new classes TaxonomicView and TaxonNode.

See taxonomic view for further information.

DESCRIPTIVE DATA - PROPOSED REVISIONS


1. MAJOR - Character/Descriptor/Feature concept

Impacted objects: Feature

The Feature class is described in the class comments by: "The class for individual properties (also designed as character, type or category) of observed phenomena able to be described or measured."

a. Issues

It is very interesting that the object Feature is not typed such as Characters in SDD (Categorical, Quantitative, etc.) or many other models. However, if the information is needed as to what kind of data is supported by a certain Feature, it is not clearly stated how to understand and use the different attributes. Moreover, there are a dozen categories of Features (Additional Publication, Image, Cultivation, Description, ...) that are rich but difficult to interpret in the case of the import.

As a reminder, below is the list of the Feature class attributes:

  • supportsTextData -> feature can be described with TextData objects

  • supportsQuantitativeData -> feature can be described with QuantitativeData objects

  • supportsDistribution -> feature can be described with Distribution objects (geographical)

  • supportsIndividualAssociation > feature can be described with IndividualsAssociation objects (between the described specimen and a second one - for instance a host, only for SpecimenDescription)

  • supportsTaxonInteraction > feature can be described with TaxonInteraction objects (between the described Taxon and a second one - for instance a parasite, a prey or a hybrid parent, only for TaxonDescription)

  • supportsCommonTaxonName -> feature can be described with CommonTaxonName objects

  • recommendedModifierEnumeration -> set of TermVocabulary containing the Modifier objects recommended to be used for DescriptionElementBase elements

  • recommendedStatisticalMeasures -> set of StatisticalMeasure recommended to be used in case of QuantitativeData

  • supportedCategoricalEnumerations -> set of TermVocabulary containing the list of possible State to be used in CategoricalData

The flexibility of the Feature class is not a problem for the import of SDD descriptive data: for each character, a new DESCRIPTION Feature instance is created:

  • for SDD CategoricalCharacter, supportedCategoricalEnumerations is set with the states defined in SDD in the elements StateDefinition

  • for SDD QuantitativeCharacter, supportsQuantitativeData is set to true.

  • for SDD TextCharacter, support supportsTextData is set to true.

  • SDD SequenceCharacter: so far, this data are not imported and I don't have an SDD example of this element being used. I guess it should be imported in a Sequence object?

However, exporting SDD data raises questions about the object Feature. I can see 3 different problems:

  1. There is no safeguard to ensure that DescriptionElementBase objects used for a description tally with the way the corresponding Feature has been described (for example, a DescriptionElementBase associated with a Feature that has only information on supportedCategoricalEnumerations, could be of the type QuantitativeData).

  2. The SDD standard and most descriptive models require the definition of a descriptive system (list of characters, potential states, potential measures) before expressing the strutured descriptions through this descriptive system. It is difficult to export properly this descriptive system to SDD: I can either export all the Feature (but most of them will be non relevant to the exported descriptions), or I can create the descriptive system by scanning all descriptions to extract only characters that are effectively used in the concerned descriptions (loss of efficiency).

  3. In SDD, categorical states do not have to be defined at the Character level, they can be defined at a more general level and shared. Therefore, the supportedCategoricalEnumerations could well be empty: how do we know then that it supports StateData?

b. Example

If we consider the feature (character/descriptor in other models) "Leaf length". Below are examples corresponding to each problem described above:

  1. A new Feature Instance names "Leaf length" is created with the attribute supportsQuantitativeData set to true and supportedCategoricalEnumerations set to null. It is still possible to create a DescriptionElementBase of type CategoricalData with the attribute feature set to "Leaf length" feature, and for example, the attribute states set to a list of StateData containing one item {"small"}. -> A feature described as a quantitative feature is used as a categorical feature.

  2. Exporting 2 descriptions from the CDM, which contain only 1 DescriptionElementBase, such as:

Viola hederacea -> Leaf Length (mm) -> {Min = 2.3, Mean = 5.1, Max = 7.9, SD = 1.3, N = 20}

Viola betonicifolia -> Leaf Length (mm) -> {Min = 2.9, Mean = 5.3, Max = 7.4, SD = 1.3, N = 21}

There might be other Feature instances stored in the CDM ("Leaf complexity", "Body shape", "Flattening of body", ...) related or not to the descriptions of such plants.

Therefore, when exporting the descriptive system, either there will be a majority of non-used features exported, if all feature are exported, or descriptions will have to be scanned one by one to detect only effectively used ones. For the last solution, it is ok with this simple example, but if with potentially hundreds of descriptions and hundreds of characters, the complexity increases quickly.

  1. The states "small", "medium", "large" could be defined as DescriptiveConcept elements in SDD and the CategoricalCharacter "Leaf length" could contain no StateDefinition elements, using the stated defined more generally in CodedDescriptions. In this case, when the character "LeafLength" is imported, a Feature with no supportedCategoricalEnumerations is created. This Feature type is undefined while it supports CategoricalData.

c. Current solution

For now, all Feature instances are exported.

d. Proposed change (NOT IMPLEMENTED)

I think there should be a distinction within Feature attributes, between the type of data supported by the Feature (supportsTextData, supportsQuantitativeData, etc.) and the domain of possible values or frame of reference (recommendedStatisticalMeasures, supportedCategoricalEnumerations).

In practical terms:

  • I would add a boolean to the attribute: 'supportsCategoricalData' (IMPLEMENTED),

  • I would remove the domain of possible values (recommendedModifierEnumeration, recommendedStatisticalMeasures, supportedCategoricalEnumerations) and create a new class that we could call for example PossibleValues or RecommendedValues from which would inherit RecommendedModifiers, RecommendedStates, and RecommendedStatisticalMeasures.

  • I would add an attribute (e.g. PossibleValuesDomains) that would be a Set).

It doesn't prevent problem 1 from happening but at least it clarifies the typing of Feature objects: it is set only through the boolean attributes 'supports...'.

It doesn't resolve problem 2. I would suggest to attach an DescriptiveSystem object to a DescriptionBase object (see item 6).

It resolves problem 3. The typing of the Feature will only depend on the boolean attributes.

Hagedorn - 27/02/2009 One comment on PossibleStatisticalMeasures: at this point both SDD and CDM take the position that all statistical measures known to the system are in principle valid data and thus allowed. At the same time, the designer of a matrix has a valid interest to make a choice of preferred measures. This is the reason why we speak of "recommendedStatisticalMeasures". Example: Leaf Length, Kurtosis = 2.3 is just as valid a statement (although highly unlikely) as Leaf Length, mean = 12.3. However: Flower color = Long is simply wrong. Thus the strict enforcement of possible states.

The base class seems reasonable, I would, however, recommend renaming it from PossibleStates to AvailableStates.

Müller - 27/02/2009 The PossibleValues class seems reasonable to me but instead of having subclasses all having the same structure we could use Java generics instead

Class PossibleValues{

Set supportedValues;

}

and/or something similar for the Vocabulary based supported values and IPossibleValue implemented by all relevant classes like MeasurementUnit and StatisticalMeasure

Fradin - 23/03/2009

The updated proposed change (NOT IMPLEMENTED) is summarized in the diagram below.

Two new classes are suggested:

  • AvailableValues: makes it possible to express a frame of reference of values (e.g. an extensive range of colors). The 4 types of values that can be listed all implement a new interface: IAvailableValue. It inherits from TermBase, so that it can be described through the attribute representations.

  • RecommendedValues: makes it possible to list the values that can be used in a certain context (e.g. for a group of taxa and a specific feature, only a limited range of colors can be used). It can reference an AvailableValues instance that corresponds to the larger frame of reference (e.g. a standard range of colors). It inherits from IdentifiableIdentity: a title could be enough to describe this specific set.


2. HIGHLY CRITICAL - Mixed properties associated with mixed objects

Impacted objects: all objects inheriting from VersionableEntity

a. Issue

[Gregor I find your observation about the limitation that "essential general properties (title, description, media and original sources) are available only for a restricted number of objects" very interesting. I had some discussions with Markus, trying to get him on erring on the side of allowing sometimes a property which is only necessary under very special use cases, rather than custom tailoring properties to the currently perceived needs. I can understand that Markus wanted to have a clean model, but since in SDD we started doing this, and in the end found that more and more things are shared, we at some point decided to move quite a bit (I am not claiming the fully correct bit) into the abstract base classes.

The "precision" aimed at, is also in my view responsible to the deep class hierarchy, which hinders a ready understanding of the model. From the UML it is difficult to derive which properties some derived classes have, because all inheritance layers contribute.

d. Proposed change (NOT IMPLEMENTED)

I think these properties should be made generic, therefore available at a higher level.

The specific attributes I am thinking of are: representations (Set), media (Set), sources (Set).

To implement this, I can see 2 solutions: a drastic one and a less drastic one.

  • drastic (directly inspired from the use of the SDD Representation element) : the problem is that it would impact the CDM at a high level so I am probably overlooking important issues raised by this.

It consists in having these attributes at the level of the VersionableEntity object. However, as the Representation, Media and OriginalSource classes all inherit from VersionableEntity, they should be removed from this hierarchy of objects and defined independantly.

The new VersionableEntity attribute would be:

representations: Set

and the Representation object, defined independantly, would contain media and sources as attributes.

In parallel, redundant attributes in lower classes could be removed.

Therefore, any CDM object inheriting from VersionabeEntity could be represented in the same way: a title and a description (possibly available in several languages), one or several images attached to the object, and one or several sources.

  • less drastic: to make available these properties largely, they could be put back up in the hierarchy.

I would suggest:

 > adding to TermBase: sources + media

 > adding to Media: representations

 > adding to ReferencedEntityBase: media

 > adding to IdentifiableEntity: representations + media

 > adding to FeatureNode: representations + media + sources

 > removing media from DefinedTermBase

 > removing media from DescriptionElementBase

 > removing media from IdentifiableMediaEntity

Müller - 27/02/2009 I clearly see your point that you needed some attributes but they were not available due to the class hierarchy. Anyway I think to have too many attributes at the very high levels makes thinks more confusing for the user sometimes and opens possibilities which should exist. This sometimes makes e.g. exporting data more difficult if you do not want to loose information. E.g. if you have representations for each versionable entity you will have to check if someone for some reason added some representation to a TaxonNameBase which is a versionable entity. This doesn’t make sense because a TaxonNameBase is always meant to be a scientific name (otherwise you should use CommonName) and therefore only Latin should be available. Also a TaxonNameBase does not really need a media, but if this possibility exists people may start to save protologues as media directly with the name instead of using a TaxonNameDescription. So you will start having the same type of information at different places and you have to check them all, if you don’t want to loose information. So I don’t think that e.g. representations should be available to each class because there are many classes that do not need them really.

Therefore I think we should keep the number of attributes as limited as possible to each but at the same time of course we need to be able to express things that have to be expressed by adding necessary attributes.

Maybe you could set up a table with all classes where you think representations, media or originalSources are really needed from your experience with SDD/CDM.

I also feel a bit uncomfortable with having media and textual representation within one class because I think many representations are more abstract so we will never need a media for it. But I can see that this way of thinking is maybe influenced by the way we use representations now and that is only for defined terms. Many of the defined terms do not seem to have a need for a media representation. If you use representation in a more general way this may change.

I know that my arguments may go against the open world assumption that is followed by the TDWG ontology for example. But from my perspective the CDM should be a DataModel that is complex but still made to be used in concrete application. Therefore it tries to be strict were ever possible. At the same time I am not sure if this is always the right way to go so I am looking forward for the discussion about the above issues.

Clark - 03/03/2009 suggested that we could make a TermBase an IdentifiableEntity - IdentifiableEntities do have a collection of OriginalSources, and space for the IdInSource.

Fradin - 24/03/2009 The diagram below represents the solution proposed by Ben, i.e. to make a TermBase an IdentifiableEntity so that they can have a collection of OriginalSources. I TESTED it in my environment and it works fine. However, I had to add the method generateTitle() to a certain number of classes. I think it still needs to be discussed because the representations attribute becomes redundant with titleCache and then we are exactly with the problem of too many attributes mentioned by Andreas.


3. MAJOR - Creation of a defined set of descriptions

Impacted objects: new object

a. Issue

Cf. mail exchanges between Gregor Hagedorn, Ben Clark and Helene Fradin in December 2009 "Keys and descriptions in the CDM".

There is no equivalent way of representing a SDD Dataset into the CDM and multi-access keys.

d. Proposed change (TESTED LOCALLY)

The solution proposed by Ben was a delimited set of taxa and their description. It would certainly be helpful for the import/export between SDD and CDM.

[Gregor] Perhaps to generalize this, a working set of taxa and a default character tree (to optionally create a subset of all taxa) could be provided? Such a working set could then carry a flag that it is suitably revised to serve as a multi-access key.

public class WorkingSet {

private Map matrix;

private DescriptiveSystem descriptiveSystem;

private boolean multiAccessKey;

private Language defaultLanguage;

...

}

Hagedorn - 27/02/2009 I wonder whether separate Working Set Taxa and Working Set Features may not be more desirable, than actually filtering on DescriptionBase objects directly. To me it certainly seems to be more logically and consistent to set up ("I work with this genus/family etc, and have selected 200 out of 900 available

features").

Filtering the matrix dimension might also help in making it more similar to SDD Dataset element.

Fradin - 24/03/2009 New proposition (TESTED LOCALLY) with 2 separate working sets classes (NOT IMPLEMENTED)

public class WorkingSetTaxa {

private Set taxa;

...

}

public class WorkingSetFeatures {

private Set descriptiveSystem;

...

}


4. MAJOR - Mapping use and referential objects


5. MAJOR - Problem how CDM handles the link between description and scientific taxonomic name

a. Issue

The fact that structured descriptions (DescriptionBase objects) cannot always be linked with a scientific taxonomic name raises problems for regrouping related descriptions. If the only possibility to regroup descriptions is by using the association with an existing taxonomic hierarchy, it limits the possibility of extracting sets of descriptions from the CDM. In addition, when importing data into the CDM, the information on potential connections between descriptions other than taxonomic is lost if not structured identically (e.g. use of the Scope class). A model such as SDD uses a Dataset object which contains a set of descriptions that can be tagged with a name, a description and media objects.

Hagedorn - 27/02/2009 About the problem how CDM handles the link between description and scientific taxonomic name I tend to agree. I have had repeated problems understanding the relation between descriptions and taxon names (and that they are so different).

I want to make sure that the use-case of unidentified specimens is included; this is possible both in CDM and in SDD (description has specimen but no taxon scope) (In fact this is simply equivalent to being identified as "Biota" and can easily be converted to that.)


6. MINOR - Descriptive system

Impacted objects: DescriptionBase

a. Issue

There is no possibility of associating a set of features/characters/descriptors to a description, or a set of descriptions.

d. Proposed change (IMPLEMENTED as an attribute of DescriptionBase)

To create a new object called DescriptiveSystem which contains at least a set of Feature objects possibly associated with domain of values.

public class DescriptiveSystem {

private Set features;

// OR private Set>;

}

CURRENT INTERMEDIARY IMPLEMENTATION: http://dev.e-taxonomy.eu/trac/attachment/wiki/CdmVersionTwoDiscussion/DescriptionBase.gatcl.PNG

Fradin - 24/03/2009 Exactly corresponds to the WorkingSet classes defined in 3.


7. MINOR - How to express uncertainty or inapplicability ?


8. MINOR - Handling of multiple languages


9. MINOR - Media properties and associations

IMPLEMENTED


10. MINOR - A default measurement unit for Feature


11. MINOR - Ordering of TermVocabulary for supportedCategoricalEnumerations in Feature


12. MINOR - Why is the setParent function not public in FeatureNode ?


13. MINOR - How to distinguish between characters and groups as they are both Feature objects ?

Should the 'partOf' attribute be used?


14. MINOR - How to export and reimport multi-types characters/features/descriptors between CDM and SDD ?


15. MAJOR - How to attribute a language to a titleCache ?

d. Proposed change (NOT IMPLEMENTED)

To make titleCache a LanguageString or a MultiLanguageText.


16. MINOR - Coding status / Unknown description

d. Proposed change (NOT IMPLEMENTED)

A new object would be created to be able to express a generic status information about a DescriptionBaseElement. For example, if a feature is “color of wings” but the described taxon has no wings, the description element associated with this feature is “not applicable”. It needs to be distinguished from a case where the description information is missing because “not available”. A DescriptionBaseElement could have a status attribute corresponding to a Status object expressing this information. This object would be similar to the DataStatus element in SDD and allow the recording of standardized reasons why data are missing. UBIF terminology (common foundation for several TDWG/GBIF standards like SDD) would be used.

[!Statusjpg!|[Gregor Hagedorn - 27/02/2009]] We did have Coding Status control at some point in the dicussions with Markus. This is a question to Andreas etc, wether perhaps it is just hidden in some way.

Otherwise I can only confirm how important the management of coding-/data-entry-status is (which also includes status values such as "Not to be coded = Decided to not code" or "Unknown = Research was performed, but still unknown". No status value is interpreted as: "work to do".

Müller - 27/02/2009 Can more then one of the values of these status appear for the same description element? If no, we should handle status as a defined term instead of having multiple boolean values.

Is the status type information enough or should we handle it the same way as e.g. NomenclaturalStatusType which inherits from ReferencedEntityBase, so you can also store the information about who thinks that this status is the valid one. Please don’t misunderstand me here. I don’t want to make it more difficult then necessary but I just made the experience that taxonomists always want to add references to any kind of information and so I wonder if this is needed here too sooner or later. But if SDD does not have it, maybe we do not need it.

Clark - 27/02/2009 I think that the idea for recording the status of an object is really important, but I wonder how this use case differs from Marker (which has a boolean, and a MarkerType which already has values like COMPLETE, DOUBTFUL, TO_BE_CHECKED).

Having multiple ways to do the same thing is not a good idea, so if we use Status to describe the status of a DescriptionElementBase, then we should not use Marker for this purpose.

Updated by Andreas Müller almost 15 years ago · 62 revisions