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'M Training A Pytorch Model And Save It Every Epoch. It Seems Like The Model Wights Are Overridden And I Can'T Choose The Best Model After The Experiment Ends. This Feature Is Missing Or I'M Not Using The Library Well?

Hi there,
I'm training a pytorch model and save it every epoch. It seems like the model wights are overridden and I can't choose the best model after the experiment ends.
This feature is missing or I'm not using the library well?

  
  
Posted 4 years ago
Votes Newest

Answers 9


SteadyFox10 ModelCheckpoint is not for pytorch I think, couldn't find anything like it.

  
  
Posted 4 years ago

Hi PompousBeetle71 I'm with SteadyFox10 on this one. Unless you choose a file name based on epoch or step , you are literally overwriting the model file, which Trains will reflect. If you use epoch in the filename you will end up with all your models logged by Trains. BTW we are actively working on integration with pytorch ignite, so if you have any suggestions now is the time :)

  
  
Posted 4 years ago

PompousBeetle71 just making sure, and changing the name solved it?

  
  
Posted 4 years ago

SteadyFox10 AgitatedDove14 Thanks, I really did change the name.

  
  
Posted 4 years ago

Well, I use ignite and trains-server with a logging similar to ignite.contrib.handlers so I will be very happy to test this integration.

  
  
Posted 4 years ago

SuccessfulKoala55 please post here once the code is available in your pytorch_ignite 🙂

  
  
Posted 4 years ago

PompousBeetle71 , check the n_saved parameter on the ModelCheckpoint creation.

  
  
Posted 4 years ago

Oh sorry, I was thinking about ignite (I don't know why) not trains. The only way I know is to use a different name when saving. I personnaly use f"{file_name}_{epoch}_{iteration}" .

  
  
Posted 4 years ago

AgitatedDove14 Yes.

  
  
Posted 4 years ago
687 Views
9 Answers
4 years ago
one year ago
Tags