Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[AzureML] ParallelRunStep multiple outputs #18440

Closed
emepetres opened this issue Apr 29, 2021 · 9 comments
Closed

[AzureML] ParallelRunStep multiple outputs #18440

emepetres opened this issue Apr 29, 2021 · 9 comments
Labels
ADO Issue is documented on MSFT ADO for internal tracking customer-reported Issues that are reported by GitHub users external to the Azure organization. Machine Learning ML-Inference AreaPath question The issue doesn't require a change to the product in order to be resolved. Most issues start as that Service Attention Workflow: This issue is responsible by Azure service team.

Comments

@emepetres
Copy link
Contributor

Is your feature request related to a problem? Please describe.
I have a batch inference pipeline of a model that outputs more than one output. I'd like ParallelRunStep could allow multiple outputs in the same way as PythonScriptStep.

Describe the solution you'd like
I want to define the output parameter of ParallelRunStep constructor as a list of PipelineData or OutputPortBinding instead of a single element.

Describe alternatives you've considered
Currently I have an output with subfolders. This is inefficient as I have to pass the whole output to next steps, that only use a part of it.
@ghost ghost added needs-triage Workflow: This is a new issue that needs to be triaged to the appropriate team. customer-reported Issues that are reported by GitHub users external to the Azure organization. question The issue doesn't require a change to the product in order to be resolved. Most issues start as that labels Apr 29, 2021
@rakshith91 rakshith91 added Machine Learning Service Attention Workflow: This issue is responsible by Azure service team. labels Apr 30, 2021
@ghost ghost removed the needs-triage Workflow: This is a new issue that needs to be triaged to the appropriate team. label Apr 30, 2021
@ghost
Copy link

ghost commented Apr 30, 2021

Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @azureml-github.

Issue Details

Is your feature request related to a problem? Please describe.
I have a batch inference pipeline of a model that outputs more than one output. I'd like ParallelRunStep could allow multiple outputs in the same way as PythonScriptStep.

Describe the solution you'd like
I want to define the output parameter of ParallelRunStep constructor as a list of PipelineData or OutputPortBinding instead of a single element.

Describe alternatives you've considered
Currently I have an output with subfolders. This is inefficient as I have to pass the whole output to next steps, that only use a part of it.

Author: emepetres
Assignees: -
Labels:

Machine Learning, Service Attention, customer-reported, needs-triage, question
Milestone: -

@rakshith91
Copy link
Contributor

Thanks for posting! We'll take a look asap

@v-strudm-msft v-strudm-msft added the ML-Inference AreaPath label Apr 30, 2021
@ghost
Copy link

ghost commented Apr 30, 2021

Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @shivanissambare.

Issue Details

Is your feature request related to a problem? Please describe.
I have a batch inference pipeline of a model that outputs more than one output. I'd like ParallelRunStep could allow multiple outputs in the same way as PythonScriptStep.

Describe the solution you'd like
I want to define the output parameter of ParallelRunStep constructor as a list of PipelineData or OutputPortBinding instead of a single element.

Describe alternatives you've considered
Currently I have an output with subfolders. This is inefficient as I have to pass the whole output to next steps, that only use a part of it.

Author: emepetres
Assignees: -
Labels:

ML-Inference, Machine Learning, Service Attention, customer-reported, question
Milestone: -

@shivanissambare
Copy link

@tracychms can you please help with this issue, thank you :)

@tracychms
Copy link

Thanks for the feedback.
Could you please clarify how you want multiple outputs? Are you looking for one scoring script run() produces one output (one PipelineData or OutputPortBinding), and only one output will be used in the next step (either ParallelRunStep or other type of step)?

@emepetres
Copy link
Contributor Author

What I'm looking for is to have a ParallelRunStep that produces two outputs (two PipelineData for example), so they can be consumed separatelly by two PythonScriptStep. Each PythonScriptStep would take only one previous output as input:
parallel_outputs

@v-strudm-msft v-strudm-msft added the ADO Issue is documented on MSFT ADO for internal tracking label May 11, 2021
@tracychms
Copy link

Thanks for the clarification.

Currently, ParallelRunStep cannot produce multiple outputs (OutputFileDatasetConfig or PipelineData) the same way as PythonScriptStep due to the difference of execution mode. PythonScriptStep executes on a single compute node, and ParallelRunStep distributes the workload across a compute cluster and executes multiple tasks (mini batches) in parallel. The results from each task (executed on different nodes) will then be combined into one output (one PipelineData or OutputDataset object).

As what you described above, one entry script produces two types of outputs (fgr & pha), the expectation is all fgr from N tasks (mini batches) will be combined into one final output (one PipelineData or OutputDataset object), and the same for pha.

@shivanissambare
Copy link

closing this issue, please reopen if you have more questions!

@shivanissambare
Copy link

#please-close

azure-sdk pushed a commit to azure-sdk/azure-sdk-for-python that referenced this issue Apr 20, 2022
@JackTn @gk-ms
@Khushboo-Baheti @Nilambari @nimaller @pr-work @matyang22 @utbarn-ms @thebenwaters @Satya-anshu @bewaterspassover @bewatersmsft
CodeGen from PR 18440 in Azure/azure-rest-api-specs
77b977e 
Azure Networking 2021-08-01 release of monthly branch (Azure#18440)

* Adds base for updating Microsoft.Network from version stable/2021-05-01 to version 2021-08-01

* Updates readme

* Updates API version in new specs and examples

* Add AppGw swagger changes for L4 proxy (Azure#17561)

* Add AppGw swagger changes for L4 proxy

* Fix Lint Errors

* fix prettier checks

* HubRoutingPreference in VirtualHub (Azure#17609)

* commit1

* commit2

Co-authored-by: Khushboo Baheti <[email protected]>

* MultipleApipa feature VpnSiteLinkConnection and  VirtualNetworkGatewayConnection (Azure#17672)

* VngConnection

* VpnSiteLinkConnection

* fixes

* fixes

* fix2

* fixes

Co-authored-by: Khushboo Baheti <[email protected]>

* Virtual Wan P2S MultiPool feature swagger changes (Azure#17620)

* Virtual Wan P2S MultiPool feature swagger changes

* Fix Swagger LintDiff errors

* Fix LintDiff errors

* Fix errors

* Fix spec

* Fix spec

* Fix spec

* Fix LintDiff errors

* Fix LintDiff errors

* Fix SDK azure-sdk-for-net generation error

* Remove suppression

* Fix errors

* Fix Lintdiff error

* Fix PrettierCheck

* changes (Azure#18002)

* Revert "changes (Azure#18002)" (Azure#18014)

This reverts commit 320ed6a6fc5a68e8af43da303f8e1caaacf24708.

* Add nic auxiliary mode (Azure#17577)

* Add nic auxiliary mode

* fix spacing

* Fixing prettier check

* Restoring package-lock file

* Restoring package json

Co-authored-by: Prachi Bhavsar <[email protected]>

* Connection Draining add new properties (Azure#18052)

* merge

* fix

* fix

* Adding express route port authorization apis (Azure#17582)

* adding apis and updating resource to support ports auth

* moving change to 2021-08-01

* minor: removing change from 2020-07-01

* lintdiff : adding type object

* minor: fixing prettier

* adding authorizations to ports property

* fixing circuitResourceUri property name

* fixing model validation

* changing circuit resource uri type to string

* removing authorizations child reosurce from parent property

* Fix Azure Firewall Policy regressions. Back fix validation issues (Azure#18233)

* Fix regressions in Firewall Policy Swagger / give firewallPolicy.json some love

* Additional lint violations

* remove breaking changes for next time. TO DO

* Revert "remove breaking changes for next time. TO DO"

This reverts commit 8f44a174c73c02d18d829f6dfb1d990488770b23.

* Reintroduce api-version for idps signature based routes. Create better names for enums to be generated in SDKs

* standardize enum names with FirewallPolicy prefix

* Azure Firewall Support of Private IP Ranges in IDPS (Azure#18320)

* Azure Firewall Support of Private IP Ranges in IDPS

* make sure all arrays have x-ms-identifiers

* FirewallPolicy not Firewall policy

fix spellcheck validation

* Ability to update tags on firewall policies (Azure#18322)

* Support updating of Azure Firewall Policy Tags. Includes HTTP Patch example

* Use common-types ErrorDetail

* Ability to update tags for Firewall Policies

* spell check fix for firewallpolicy

* Use future release api-version for example

* Added flush conn to nsg (Azure#18393)

* Added flush conn to nsg

* Updated flushConn to correct location

* Updated description

* Modified T/F to Enabled/Disabled

* Refactoring so that null value appears first

* Reverted FlushConnection to boolean value instead of string

* Revert "Added flush conn to nsg (Azure#18393)" (Azure#18576)

This reverts commit 6541d305880d1cf580496adc01f55197a01e992c.

* Fixing typo in response of idps private ip range feature (Azure#18574)

* Use common-type api version (Azure#18729)

Co-authored-by: Ben Eshed <[email protected]>

* fix (Azure#18417)

Co-authored-by: Tianen <[email protected]>
Co-authored-by: gk-ms <[email protected]>
Co-authored-by: Khushboo Baheti <[email protected]>
Co-authored-by: Khushboo Baheti <[email protected]>
Co-authored-by: Nilambari <[email protected]>
Co-authored-by: nimaller <[email protected]>
Co-authored-by: pracsb <[email protected]>
Co-authored-by: Prachi Bhavsar <[email protected]>
Co-authored-by: Matthew Yang <[email protected]>
Co-authored-by: utbarn-ms <[email protected]>
Co-authored-by: Ben Eshed <[email protected]>
Co-authored-by: Satya-anshu <[email protected]>
Co-authored-by: bewaterspassover <[email protected]>
Co-authored-by: Ben Eshed <[email protected]>
@github-actions github-actions bot locked and limited conversation to collaborators Apr 12, 2023
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Assignees
No one assigned
Labels
ADO Issue is documented on MSFT ADO for internal tracking customer-reported Issues that are reported by GitHub users external to the Azure organization. Machine Learning ML-Inference AreaPath question The issue doesn't require a change to the product in order to be resolved. Most issues start as that Service Attention Workflow: This issue is responsible by Azure service team.
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
6 participants
@emepetres @rakshith91 @paulshealy1 @tracychms @shivanissambare @v-strudm-msft