I understand that instead of having one dataset with multiple files you are planning to use existing datasets and combine them in a ‘collection’. My concerns are:
1. Our data consists of 200-8000 files, can you imagine how many datasets we’ll end up with? It will be a mess.
2. All these files in a dataset belong to each other and it doesn’t make much sense to keep them separately.
3. For performance reasons, all these files are located in a single directory which makes it easier to iterate over.
4. From my point of view, it makes perfect sense to have a concept of a “dataset” with multiple files, you have already a dataset_xxx_files folder anyway, and it’s not a big change comparing to the new concept of “collection”
5. We are already using the “m:xxx” type datasets (thanks John) in our project, I guess you don’t even have a timeframe for implementing the “collection” concept? I’m sure that for many projects using multi file datasets is a requirement now, not in ‘years’ time.
6. “Collection” is also a good idea and I guess they both can exist together, but only in the future, given current users an opportunity to use Galaxy for their needs. Otherwise we simply have to look at other frameworks which already support multi file datasets.
To reiterate what Jeremy has already said on the mailing list, this is definitely something we want, and need, for Galaxy. While this particular implementation has a lot of good parts, creating these collections as first-class composite datasets isn't ideal and we'd be stuck supporting them going forward, forever.
There's a clear plan for implementing this in Trello (https://trello.com/c/325AXIEr), most of which is straightforward to implement. The 'hard' part is really going to be implementing an ideal UI for dealing with these collections, something which we could do in phases.
What exactly are your concerns with the implementation as set out in the Trello card?
On Mon, Mar 4, 2013 at 1:32 AM, <Alex.Khassapov@csiro.au> wrote:
This is sad, I don't understand why it is such a problem? If it's already implemented and used in real projects like ours - then it is needed for the community. I don't think we have other options for our requirements, your multiple file datasets implementation was a real saviour for us.
From: email@example.com [mailto:firstname.lastname@example.org] On Behalf Of John Chilton
Sent: Monday, 4 March 2013 4:42 PM
To: Khassapov, Alex (CSIRO IM&T, Clayton)
Subject: Re: [galaxy-dev] Composite datatype output for Cuffdiff
Thanks for the comments. The galaxy team has made it clear here and to me privately that this will NOT be included in the Galaxy main code base. I hope and am I confident that they will make grouping datasets work, hopefully even to thousands of files.
I do not believe the two ideas are mutually exclusive and I will be maintaining a fork of galaxy-central with these additions, I will set this up this week hopefully. I will do my best to respond to support requests and make multiple file datasets and composite types in general as robust as possible, keep up with Galaxy updates, etc....
Obviously, it is risky to let a code base drift so far from galaxy main's however and you, me, and others who might want to use them will have to carefully weigh the risks when determining if multiple file datasets are worth the headache.
Thanks for all your help and inputs. I am sorry this did not turn out differently, I feel I have really failed here.
On Sun, Mar 3, 2013 at 10:08 PM, <Alex.Khassapov@csiro.au> wrote:
> Hi John,
> Are you saying that "composite multiple file dataset" isn't required and won't be implemented?
> We are using your implementation of multifiles dataset ("m:xxx" type) and hope that eventually it will be pushed into main Galaxy implementation.
> As we are using Galaxy for CT reconstruction tools, where input and output can consist of a couple thousand files, other options are not feasible, i.e. grouping datasets.
> -----Original Message-----
> From: email@example.com
> [mailto:firstname.lastname@example.org] On Behalf Of John Chilton
> Sent: Thursday, 28 February 2013 2:06 AM
> To: Jeremy Goecks
> Cc: Jim Johnson; <email@example.com>
> Subject: Re: [galaxy-dev] Composite datatype output for Cuffdiff
> Hey Jeremy,
> I am trying to think about a path forward with this composite multiple file dataset implementation. It seems there is consensus among the galaxy team that it shouldn't be included because grouping actual datasets would be superior. In that light, I am revisiting this e-mail, because depending on the implementation of what you described multiple file datasets are a specific case of this concept with some likely uncontroversial enhancements for the specific case of composite datatypes that are a homogeneous list of files. Does that make any sense?
> If I implemented (i) and (ii) in such a way that the multiple file dataset stuff flowed out more organically is there any chance than it could be included in galaxy-central. If no and the implicit datatypes and parallelism stuff would remain no-gos implementing what you described would still benefit the multiple file datasets implementation, so I still might do this, would a clean implementation of just what you described be accepted?
> Any thoughts you or anyone has on the future direction of composite datatypes in general would be appreciated?
> Thanks for your time,
> On Fri, Oct 12, 2012 at 2:44 PM, Jeremy Goecks <firstname.lastname@example.org> wrote:
>> Hi Jim,
>> This is nice and is a path forward for the immediate future.
>> That said, a couple extensions to Galaxy to better support composite
>> datatypes would enable cummerbund without the additional tools:
>> (i) extending the composite datatype to include definition of
>> individual outputs in the collection;
>> (ii) extend the history panel to allow usage/selection of (1) the
>> complete composite set of files or (2) individual items in a
>> composite datatype
>> Once (i) is done, (ii) should be straightforward using the new
>> history panel code.
>> Of course, the advantage of these extensions is that they'd address
>> both cummerbund issues as well as other challenges, such as using
>> output from the barcode splitter.
>> On Oct 11, 2012, at 6:14 PM, Jim Johnson wrote:
>> Checking to see if there is any interest in including a parameter
>> option to select outputs for cuffdiff, potentially including a
>> composite output and a cummeRbund sqlite database.
>> cuffdiff produces 21 output files, which is a little unwieldy in a
>> galaxy history.
>> cummeRbund generates its database when given a cuffdiff output
>> directory, but manually hooking up 21 outputs to the cummerbund_wrapper is a pain.
>> I've put demo code in the testtoolshed under the name repository name
>> This includes new datatypes defined in datatypes_conf.xml and
>> implemented in
>> <!-- html composite dataset with cuffdiff outputs in the extra
>> files path -->
>> <datatype extension="cuffdata"
>> <!-- cummeRbund SQLite database -->
>> <datatype extension="cuffdatadb"
>> The cuffdiff wrapper has a multiple select parameter to choose which
>> output files to put in the history.
>> In addition to the 21 cuffdiff outputs, the wrapper can also generate:
>> cuffdata - which is a composite HTML output with links to the 21
>> cuffdiff outputs
>> cuffdatadb - which is the cummeRbund SQLite database
>> I also added utility tools:
>> cuffdata_datasets - which will take files from the composite
>> cuffdata and copy them as datasets into the history
>> cuffdata_cummerbund - which generates the cummeRbund cuffdatadb
>> from the composite cuffdata
>> I updated the cummerbund_wrapper:
>> with tryCatch so that a R error on a plot won't exit the Rscript
>> to include a small image of each plot on the html page
>> added plots for : dispersion, scatter matrix, MDS, and PCA
>> Please keep all replies on the list by using "reply all"
>> in your mail client. To manage your subscriptions to this and other
>> Galaxy lists, please use the interface at:
> Please keep all replies on the list by using "reply all"
> in your mail client. To manage your subscriptions to this and other Galaxy lists, please use the interface at:
Please keep all replies on the list by using "reply all"
in your mail client. To manage your subscriptions to this
and other Galaxy lists, please use the interface at: