Skip to content
Snippets Groups Projects

Legacy algorithm warning

Merged Samuel GAIST requested to merge legacy_algorithm_warning into 1.5.x
All threads resolved!

This merge request adds a warning when running a legacy algorithm.

They should be ported to the new API.

Merge request reports

Pipeline #20452 passed

Pipeline passed for f1680a23 on legacy_algorithm_warning

Test coverage 71.00% (0.00%) from 4 jobs

Merged by André AnjosAndré Anjos 7 years ago (May 23, 2018 7:40pm UTC)

Loading

Pipeline #20454 passed

Pipeline passed for 2f5ba8c6 on 1.5.x

Test coverage 71.00% (0.00%) from 4 jobs

Activity

Filter activity
  • Approvals
  • Assignees & reviewers
  • Comments (from bots)
  • Comments (from users)
  • Commits & branches
  • Edits
  • Labels
  • Lock status
  • Mentions
  • Merge request status
  • Tracking
  • Apart from the comment above, LGTM. Thanks for this!

  • Samuel GAIST added 1 commit

    added 1 commit

    • f1680a23 - [executor] Log warning when using a legacy algorithm

    Compare with previous version

  • Samuel GAIST resolved all discussions

    resolved all discussions

  • Thanks - great job!

  • merged

  • André Anjos mentioned in commit 2f5ba8c6

    mentioned in commit 2f5ba8c6

  • @samuel.gaist what exactly does the legacy API entail?

    We really need documentation on the API because right now there's nothing.

    I'd like to avoid the legacy API for my example code in the tutorial but I need to know what that means.

  • The Algorithm and related classes are documented in the code files. If nothing appears in the documentation then it's likely a bug in the documentation generation step.

    The new API changes the way the algorithm is initialised as well as the access to the data. This has been made to be able to use GPUs which are not using sequential data access (they could but that's just killing any benefits to use GPUs since they would have to wait most of the time to only get one piece of data at a time).

    You now have a setup step that allows the user to e.g. load models, prepare data, etc. Basically, things that need to be done only once (which was not possible with V1 where you had to do that within the processing itself).

    Then you have the processing that can either be done sequentially (one input data after the other) or autonomously (i.e. the GPU grabs the data it needs when it wants).

    You can also find how they are being used in the test_algorithm.py file.

  • Please register or sign in to reply
    Loading