dbispipeline issueshttps://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues2022-05-02T13:33:29+02:00https://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/64Storing results in database crashes when using a feature union2022-05-02T13:33:29+02:00Manfred MoosleitnerStoring results in database crashes when using a feature unionWhen the dbispipeline stores the results, the pipeline configuration is extracted from the pipeline object. The code differentiates three cases, where the last case is for when a "transformer_list" is available, which is the case, e.g., ...When the dbispipeline stores the results, the pipeline configuration is extracted from the pipeline object. The code differentiates three cases, where the last case is for when a "transformer_list" is available, which is the case, e.g., when a feature union is used. The code then tries to extract the individual transformers and expects the "transformer_list" to be of type dictionary, as a method "items()" is called, but as the name suggests, that variable is of type list when using "sklearn.pipeline.FeatureUnion".Manfred MoosleitnerManfred Moosleitnerhttps://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/62Discussion: easier replicability for external users?2021-03-25T16:02:40+01:00User expiredDiscussion: easier replicability for external users?for a recent paper, I made a repository public that contained the code for my experiments.
Thereby, I noticed that it is not very easy or intuitive to do so: external users have to setup a database, configure the ini file, ...
While som...for a recent paper, I made a repository public that contained the code for my experiments.
Thereby, I noticed that it is not very easy or intuitive to do so: external users have to setup a database, configure the ini file, ...
While some of these issues can be circumvented by the project (e.g., providing a requirements.txt so that the user can decide whether a virtual env is used or not), what could be a viable strategy to enable quick execution of the experiments?
I figure that using the `--dryrun` flag is one quick solution where the user does not need to setup a database :thinking:
This ticket is meant to be a brainstorming area.https://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/35Include execution time2020-05-05T13:50:32+02:00User expiredmichael.voetter@uibk.ac.atInclude execution timeIt would be handsome if we store the total execution time in the database.It would be handsome if we store the total execution time in the database.https://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/32Remove old checkpoints2020-04-09T16:46:52+02:00User expiredmichael.voetter@uibk.ac.atRemove old checkpointsWrite tooling to clean up and remove old (unused) checkpoints.
Store useful meta data for that.
Contributes to #22Write tooling to clean up and remove old (unused) checkpoints.
Store useful meta data for that.
Contributes to #22https://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/31Checkpointing in evaluators2020-04-09T16:45:19+02:00User expiredmichael.voetter@uibk.ac.atCheckpointing in evaluatorsEvaluator coordinates checkpointing.
* time based
* step based
Try loading a checkpoint before starting the evaluation/experiment.
Model triggers checkpoint request. This request is pushed to the upper most evaluator. This root evalua...Evaluator coordinates checkpointing.
* time based
* step based
Try loading a checkpoint before starting the evaluation/experiment.
Model triggers checkpoint request. This request is pushed to the upper most evaluator. This root evaluator decides if a checkpoint is created and delegates the creation of the checkpoint. Pickle dump as fallback try delegation first.
Contributes to #22https://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/30Checkoint models2020-04-09T16:41:32+02:00User expiredmichael.voetter@uibk.ac.atCheckoint modelsHow to checkpoint models?
* call `create_checkpoint` and `restore_checkpoint`
Contributes to #22How to checkpoint models?
* call `create_checkpoint` and `restore_checkpoint`
Contributes to #22https://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/29Checkpoint data2020-04-09T16:39:49+02:00User expiredmichael.voetter@uibk.ac.atCheckpoint dataHow to checkpoint data?
* dataloader (ensure that we get the same data)
* datashapes etvl. checksums
* splits
Contributes to #22How to checkpoint data?
* dataloader (ensure that we get the same data)
* datashapes etvl. checksums
* splits
Contributes to #22https://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/26Extend the documentation of plans2020-04-09T11:27:08+02:00User expiredmichael.voetter@uibk.ac.atExtend the documentation of planshttps://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/24Extend cli documentation.2020-04-09T11:00:27+02:00User expiredmichael.voetter@uibk.ac.atExtend cli documentation.We should add some documentation on the features of the CLI in the README.
The following discussion from !10 should be addressed:
- [ ] @c7031257 started a [discussion](https://git.uibk.ac.at/dbis/software/dbispipeline/merge_requests/1...We should add some documentation on the features of the CLI in the README.
The following discussion from !10 should be addressed:
- [ ] @c7031257 started a [discussion](https://git.uibk.ac.at/dbis/software/dbispipeline/merge_requests/10#note_58472): (+1 comment)
> We should write some documentation on these options. E.g. in the README but this can be deferred.
>
> Maybe it is better to open an issue for that?https://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/22Checkpoint feature2020-04-09T16:46:52+02:00User expiredmichael.voetter@uibk.ac.atCheckpoint featureImplement a checkpoint feature that allows to continue computation.
Possible things to consider:
* a cluster automatically restarts a job -> automatic detection of a checkpoint and start there
* but the same job can be run multiple time...Implement a checkpoint feature that allows to continue computation.
Possible things to consider:
* a cluster automatically restarts a job -> automatic detection of a checkpoint and start there
* but the same job can be run multiple times
* we need to include intermediate results and data in addition to the models/pipeline
* only some models/pipeline configurations allow checkpoints
Maybe we can reuse the `backup_dir` of #9.
Related to #4 and #15.https://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/15Persist Models2020-09-17T08:50:06+02:00User expiredmichael.voetter@uibk.ac.atPersist ModelsImplement model storing and loading.Implement model storing and loading.https://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/14Fix code formating2019-11-27T09:40:42+01:00User expiredmichael.voetter@uibk.ac.atFix code formatinghttps://git.uibk.ac.at/informatik/dbis/software/dbispipeline/-/issues/4Implement possiblility to re-run based on db2020-04-06T11:04:10+02:00User expiredmichael.voetter@uibk.ac.atImplement possiblility to re-run based on dbNeeds #2 to be fixedNeeds #2 to be fixed