Examples: query, "exact match", wildcard*, wild?ard, wild*rd
Fuzzy search: cake~ (finds cakes, bake)
Term boost: "red velvet"^4, chocolate^2
Field grouping: tags:(+work -"fun-stuff")
Escaping: Escape characters +-&|!(){}[]^"~*?:\ with \, e.g. \+
Range search: properties.timestamp:[1587729413488 TO *] (inclusive), properties.title:{A TO Z}(excluding A and Z)
Combinations: chocolate AND vanilla, chocolate OR vanilla, (chocolate OR vanilla) NOT "vanilla pudding"
Field search: properties.title:"The Title" AND text
Hi Guys, Just Downloaded Clearml. Wonder If I Can Manage Versions, Lets Say Some Hyperparameters "Try1" With Data "Dat1" And A Model Will Be One Version And "Try2" With Slight Change To Dat1 With Same Model Will Be Second Version, Where I Can Go Bacl In

Hi guys, just downloaded clearml. wonder if I can manage versions, lets say some hyperparameters "try1" with data "dat1" and a model will be one version and "try2" with slight change to dat1 with same model will be second version, where I can go bacl in vesions.

Posted 2 years ago
Votes Newest

Answers 7

Hi KindBlackbird59

You can always clone the first task and change the parameters in the second one, is this what you are looking for?

Posted 2 years ago

Im talking about features that DVC allows

Posted 2 years ago

Hi Alon,

Posted 2 years ago

You can compare, clone and re-run with changes every task in the system, with control on the version and the parameters

Posted 2 years ago

KindBlackbird59 I think you are looking for something like a "git repository" (which, IIRC, is how dvc sees "projects" or models),
that gives you a clear lineage (This model came first, then I got this model with this code and this data).
The way ClearML works is slightly different, and each "repo" is shown as a project which is flat. The way we envision users marking things like "model V2" is by adding tags.
The reason behind this design is that git, while has clear lineage, is harder to work when things are not linear. Yes you can start branching off master, but it gets whacky once you want to go back 2 months and find the "good" model or when you ran an HPO, found 2-3 good examples then want to branch off from them. Not a DVC expert but it's hard to ask git "give me the best model from 2 months ago in branch X" (With ClearML's queries, it is possible 😄 ).

For sure you can build lineage yourself (Every model has an associated Task that created it so you can trace these back). But we recommend tagging the important models (probably like a dvc commit).

For sure, like alon, you can use our compare and clone features to weed out the best models.

Hope I understood your point and it clarifies how we do things!

Posted 2 years ago

yes it is thank you very much!

Posted 2 years ago

great! thx:)

Posted 2 years ago
7 Answers
2 years ago
4 months ago