Hello All! I'M Writing To Express My Gratitude For Supporting Such A Great Platform As Clearml, Which I'Ve Been Actively Using Lately. I'M Running Experiments Based On Yolov5, And This Code Is Automatically Sending Various Information To The Clearml Serv

Answered

Hello all!

I'm writing to express my gratitude for supporting such a great platform as ClearML, which I've been actively using lately. I'm running experiments based on YOLOv5, and this code is automatically sending various information to the ClearML server. However, as I conduct various experiments, the capacity is filling up quickly, and I'm wondering how to solve this problem without investing money. I have a few questions regarding this issue.

Firstly, I would like to know if there is a function to delete and reload each experiment. I've noticed that I can clear each experiment through the reset button, but I don't want to clear important experiments that I may need to reload later. Is there a way to reload experiments that have been cleared once? Or is there a way to just reduce the capacity without making it clear?
Secondly, I tried to delete a project including multiple experiments, but I received the message "UNABLE TO DELETE PROJECT." The full message read, "You cannot delete project "train_GT_b64" with un-archived experiments, dataviews or models. You have 16 un-archived experiments in this project. If you wish to delete this project, you must first archive, delete, or move these items to another project." I'm not sure how to proceed in this case. I didn't do achieve for each experiment, just clear. Do I need something more?
Thirdly, I'm wondering how to reduce the size of each experiment. Currently, the YOLOv5 default setting is sending much more information to ClearML, which seems to be causing the capacity to fill up quickly. Do you have any tips on how to make the size of a single experiment smaller? For example, currently, I'm logging all of the information every iteration, is it better to log some information rather than all information? Also, I'm saving checkpoints at regular intervals, but does that count toward the capacity limit? It would be easy to reduce that if I could see which logs were taking up a lot of space in one experiment, but I don't know where to find out.
Lastly, I'm not sure what features like "archive, publish, enqueue" mean or how to use them, can I get some information on that?
Thank you again for supporting such an excellent platform. I hope to continue using it well in the future.

  				
Posted 
	one year ago

					More
				  		
  Report
		
					ZippyCamel28
				
					0

Votes Newest

Answers

Hi @<1547752799075307520:profile|ZippyCamel28> , to address your points

What do you mean by 'reload'?
You need to go into the project and archive the experiments in order to delete the project + experiments in the archive
There are some configurations you can play with to report 'less' metrics. For example sdk.metrics.plot_max_num_digits You should read here - None . To get an idea of the size of an experiment think of an experiment as a document in a backend with some connections to the metrics/logs. Basically, the more you report the 'heavier' the experiment
They are basically for your own management. Archived experiments can be deleted and this is basically to hide experiments. Publishing an experiment 'locks' it and it no longer can be reset. Enqueue basically pushes and experiment into a queue, you should read about our orchestration - None
In general I would suggest you try to host your own server. When you run the server yourself, you aren't limited by anything specific. If you work alone, you can simply run the server on your development machine and work with that 🙂

  				
Posted 
	one year ago

					More
				  		
  Report
		
					CostlyOstrich36
				
					0

Write your answer

1K Views

1 Answer

one year ago