google-code-export / dkpro-tc

Automatically exported from code.google.com/p/dkpro-tc
Other
1 stars 0 forks source link

ValidityCheckTask prevents experiments to make advantage of the USE_EXISTING policy #136

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago
The ValidityCheckTask accesses several discriminators which are actually used 
in ExtractFeaturesTask or TestTask to verify experiment setups. It has to be 
re-run when any of these discriminators changes and consequently makes the 
entire pipeline re-run, since all other task depend on it.

To change this:
- enforce running ValidityCheckTask as initial task, but make the subsequent 
tasks independent from ValidityCheckTask (ValidityCheckTask has no output 
anyways).
(not sure whether this is necessary after the first step: - instead of 
accessing the parameter values which need to be checked as task discriminators, 
access them via the TaskContext)

Original issue reported on code.google.com by daxenber...@gmail.com on 4 Jun 2014 at 4:33

GoogleCodeExporter commented 9 years ago
alternatives to accessing parameters are:

- @Attribute on class fields (like @Discriminator, but does not affect 
dependency resolution)
- implement ConfigurationAware (also does not affect dependency resolution)

Original comment by richard.eckart on 4 Jun 2014 at 4:56

GoogleCodeExporter commented 9 years ago
I have implemented a feature request in the Lab to try executing the tasks in 
the order they were added (unless mandated otherwise by data dependencies): 
https://code.google.com/p/dkpro-lab/issues/detail?id=38

Adding the ValidityCheckTask as the very first subtasks and removing the 
current dummy imports related to the ValidityCheckTask should make sure the 
task is always executed first.

Can you please test?

Original comment by richard.eckart on 15 Jun 2014 at 5:49

GoogleCodeExporter commented 9 years ago
This issue was closed by revision r896.

Original comment by daxenber...@gmail.com on 16 Jun 2014 at 10:21

GoogleCodeExporter commented 9 years ago
This issue was updated by revision r896.

tested https://code.google.com/p/dkpro-lab/issues/detail?id=38

Seems to work. Parameter sweeping with with different feature sets (i.e. 
re-running ExtractFeaturesTask) does not enforce running another preprocessing 
task. Formerly, it did.

Original comment by daxenber...@gmail.com on 16 Jun 2014 at 10:21

GoogleCodeExporter commented 9 years ago
So can we now completely remove the ValidityCheckTask.DUMMY_KEY now? It still 
seems to be used in some places. 

Original comment by richard.eckart on 16 Jun 2014 at 9:33

GoogleCodeExporter commented 9 years ago
This issue was closed by revision r898.

Original comment by daxenber...@gmail.com on 17 Jun 2014 at 9:23

GoogleCodeExporter commented 9 years ago
This issue was updated by revision r898.

Yes, it can be removed. I missed the Mallet module when updating. It's all gone 
now.

Original comment by daxenber...@gmail.com on 17 Jun 2014 at 9:23