Fwd: [OTDev] Errors and warnings in datasets
Egon Willighagen
the below post is from the OpenTox mailing list. Is there something in
CHEMINF to handle error and warning messages returned by an algorithm?
Egon
---------- Forwarded message ----------
From: Christoph Helma <he...@in-silico.ch>
Date: Mon, Oct 11, 2010 at 6:51 PM
Subject: Re: [OTDev] Errors and warnings in datasets
To: development <devel...@opentox.org>
Excerpts from Nina Jeliazkova's message of Mon Oct 11 17:27:14 +0200 2010:
> Christoph,
>
> Could you tell what kind of errors (parsing of SMILES ?) would you like to
> store into metadata? Is it possible to provide examples?
It will be a mixed bag of SMILES errors, duplicated structures,
incorrect activity entries, .... Examples can be found e.g. at
http://toxcreate.org/models under "Warnings: show". Simple string
annotation for concatenated error/warning messages could be sufficient.
> There is a proposal for errors representation
> http://opentox.org/dev/apis/api-1.1/Error%20Reports , but the objective was
> to report errors from processing services ( algorithms and models) and I am
> not quite sure it is applicable.
Yes, this is a different story. During dataset parsing I try
to parse as much information as possible and to present errors/warnings
later.
Best regards,
Christoph
_______________________________________________
Development mailing list
Devel...@opentox.org
http://www.opentox.org/mailman/listinfo/development
--
Dr E.L. Willighagen
Homepage: http://egonw.github.com/
LinkedIn: http://se.linkedin.com/in/egonw
Blog: http://chem-bla-ics.blogspot.com/
PubList: http://www.citeulike.org/user/egonw/tag/papers
Janna Hastings
sorry so long to reply -- have been travelling. Are you still around on campus at the moment?
We don't have error and warning messages modelled in CHEMINF at the moment, but it is a very good idea to add this. At the moment we have
<algorithm> has_specified_output some <descriptor>
We can modify this to have
<algorithm> has_specified_output some (<descriptor> or <error> or <warning>)
Then we need to place <error> and <warning> in the class hierarchy somewhere. Of course, they are information entities too ;-) but not *chemical* information entities. Somewhere parallel to <parameter>?
Thoughts?
Cheers, Janna
Michel Dumontier
<janna.h...@gmail.com> wrote:
> Hi Egon,
>
> sorry so long to reply -- have been travelling. Are you still around on
> campus at the moment?
>
> We don't have error and warning messages modelled in CHEMINF at the moment,
> but it is a very good idea to add this. At the moment we have
>
> <algorithm> has_specified_output some <descriptor>
>
> We can modify this to have
>
> <algorithm> has_specified_output some (<descriptor> or <error> or
> <warning>)
>
> Then we need to place <error> and <warning> in the class hierarchy
> somewhere. Of course, they are information entities too ;-) but not
> *chemical* information entities. Somewhere parallel to <parameter>?
>
> Thoughts?
Yes, absolutely - we can start developing these...
m.
--
Michel Dumontier
Associate Professor of Bioinformatics
Carleton University
http://dumontierlab.com
Janna Hastings
I am working on this change in the ontology. Now that I am looking at it, I am wondering whether the <algorithm> has error or warning messages, or the <implementation>? We have both the algorithm:
<algorithm> has_specified_output some <descriptor>
and the implementation:
<software module to calculate a chemical descriptor> conforms_to some <algorithm>, has_output some <chemical descriptor>
and then particular executions of the implementation:
<parameterized software execution> has_input some <data item>
What I'm thinking is to introduce the following:
Algorithms should have specified outputs, yes. But not inputs or error messages, since these can vary per implementation. (yes?)
Software methods should have specified inputs and outputs, and warning and error messages.
Software modules should be composed of one or more software methods.
Parameter is a role played by a data item just when it forms the input to a software method.
Message is a special kind of output from a software method, which has subtypes error and warning.
If there are no objections I will introduce these changes later today and modify the documentation accordingly. Michel do you have the latest version?
Cheers, Janna
Michel_Dumontier
From:
cheminf-...@googlegroups.com [mailto:cheminf-...@googlegroups.com] On
Behalf Of Janna Hastings
Sent: Monday, October 18, 2010 11:00 AM
To: cheminf-...@googlegroups.com
Cc: Nina Jeliazkova; he...@in-silico.ch
Subject: Re: [OTDev] Errors and warnings in datasets
Dear all,
I am working on this change in the ontology. Now that I am looking at it, I am
wondering whether the <algorithm> has error or warning messages, or the
<implementation>? We have both the algorithm:
<algorithm> has_specified_output some <descriptor>
and the implementation:
<software module to calculate a chemical descriptor> conforms_to some
<algorithm>, has_output some <chemical descriptor>
and then particular executions of the implementation:
<parameterized software execution> has_input some <data item>
What I'm thinking is to introduce the following:
Algorithms should have specified outputs, yes. But not inputs or error
messages, since these can vary per implementation. (yes?)
Yes.
Software methods should have specified inputs and outputs, and warning
and error messages.
yes
Software modules should be composed of one or more software methods.
yes
Parameter is a role played by a data item just when it forms the input
to a software method.
yup
Message is a special kind of output from a software method, which has
subtypes error and warning.
How about:
message
- Warning
o error
§ fatal error
If there are no objections I will introduce these changes later today and
modify the documentation accordingly. Michel do you have the latest version?
Should be checked in to SVN.
Frank Gibson
Michel_Dumontier
I’ve raised these issues with James, and I know he’s meaning to pick this up again soon
m.
From: cheminf-...@googlegroups.com [mailto:cheminf-...@googlegroups.com] On Behalf Of Frank Gibson
Sent: Monday, October 18, 2010 11:37 AM
To: cheminf-...@googlegroups.com
Cc: Nina Jeliazkova; he...@in-silico.ch
Egon Willighagen
<janna.h...@gmail.com> wrote:
> <algorithm> has_specified_output some (<descriptor> or <error> or
> <warning>)
>
> Then we need to place <error> and <warning> in the class hierarchy
> somewhere. Of course, they are information entities too ;-) but not
> *chemical* information entities. Somewhere parallel to <parameter>?
That looks pretty neat... it also quite well matches practices in the CDK...
Egon
--
Dr E.L. Willighagen
Postdoctoral Research Associate
University of Cambridge