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
Answered
Hi There! I Hope Everyone Is Fine. I'M Starting With Clear-Ml And I Don'T Undestand One Case. I Was Wondering If Someone Could Help Me, Thanks In Advance :) I'M Executing The Same Train Multiple Times Autologging The Model With The Joblib-Scikit Way. Aft

Hi there! I hope everyone is fine. I'm starting with clear-ml and I don't undestand one case. I was wondering if someone could help me, thanks in advance :)

I'm executing the same train multiple times autologging the model with the joblib-scikit way. After the first task was completed, I published the task and the model through the UI. Then I execute again the same py file, it creates a new task, but overrides the model I've published. The log is something like:

INFO - Found existing registered model id=c5a1e8897f6a4067a999c895436e1dac [some_route/my_model.pickle] reusing it.I was expecting that the default behaviour doesn't overrides de model, maybe I have missundestood the documentation... Could you help me?

PD: Amazing work ClearML guys :D

  
  
Posted 2 years ago
Votes Newest

Answers 3


Hi CheekyDove44

PD: Amazing work ClearML guys :D

🙏 😍

I was expecting that the default behaviour doesn't overrides de model, maybe I have missundestood the documentation

What's happening, I think, is that when creating an Input Model entity for your experiment, ClearML detects an existing model entity using the same exact weights file, and simply reuses the existing model entity (since it's an Input model which is not generated by the experiment, only used as a starting reference)

  
  
Posted 2 years ago

Thank you very much for your fast response! Digging a little in the source code, I find out which was my mistake. I haven't put anything in the output_uri parameter so I was using my local laptop as output route for saving the model. I guess that because I'm overriding the old model when I do the joblib serialization, the comparison between weights will be always "equal". Maybe this comparation should be before? Anyway It's caused because I have used always the same output path in local...

  
  
Posted 2 years ago

Perhaps it's also related to the fact the model weights file name is always identical?

  
  
Posted 2 years ago