Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Lack of explanation of difference between data that describes models #477

Open
Make42 opened this issue Feb 16, 2023 · 0 comments
Open

Lack of explanation of difference between data that describes models #477

Make42 opened this issue Feb 16, 2023 · 0 comments

Comments

@Make42
Copy link
Contributor

Make42 commented Feb 16, 2023

It is unclear what the differences are conceptually between the following things.

In the web app there are multiple types of descriptive data available for models

  • tag
  • model configuration (in the tab "network")
  • metadata

Those can also be set and gotten programmatically.

Conceptually they are all descriptive data, but it is not clear to me what the differences conceptually are. Here are some ideas of mine:

  • Model configuration contains values that are available before the training, e.g., hyperparameters, paths where the model shall be saved to, etc.
  • metadata contains values that are available after training, e.g., how many training iterations where done, how long training took, how often a certain function was called, performance values (like accuracy)
  • tags are values that are implicitly boolean (either the model has the tag or does not). These are values can be changed after the model has been uploaded.

Question 1: How are they conceptually different?

Maybe they are not supposed to be conceptually different, but just technologically. For example, you have the two methods connect and connect_configuration of Task, which do not differ in the conceptual content they contain, but in that the first connects Python symbols/variables to the task and the second connects files to the task.

Question 2: How are they technically different?

In practice, the reason why I populate those data fields before/during/after training is so that I can use them as query arguments, when looking for appropriate models when I want to use one for inference later. One of the most important query arguments here is the hyperparameters. Assuming my understanding above was correct, I do not see how I can include the model configuration into my query...!

Also, I just realized, all of those values can be changed after the model training.
It seem now to me that the "model configuration" is basically a text file and the "metadata" is a key-value story.
So maybe there is not conceptual difference at all, but just the technical difference that they are stored in a different way.
But then why call them "model configuration" and "metadata"?


If someone explains those things here, I am happy to make a pull request.

@Make42 Make42 changed the title Lack of explanation of difference between data descriping models Lack of explanation of difference between data that descripes models Feb 16, 2023
@Make42 Make42 changed the title Lack of explanation of difference between data that descripes models Lack of explanation of difference between data that describes models Feb 16, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant