search

JDACS4C-IMPROVE / IMPROVE

Libraries and scripts for basic IMPROVE functionalities
MIT License
1 stars 3 forks source link

Add a section for global parameters #55

Open jonesse3 opened 3 weeks ago

jonesse3 commented 3 weeks ago

The change can occur on line 33 in config.py per @wilke. Options are DEFAULT and GLOBAL. Vote to use GLOBAL as that makes more sense to me that it's applied everywhere. What do you think @adpartin? Once this has been fixed, I can add it to LGBM/framework-api and check that it's read.

jonesse3 commented 3 weeks ago

Once we fix this issue, I can check with LGBM.

In particular, I'll check this:

model_file_name = model model_file_format = .txt

Per @nkoussa, this has to be repeated in both Train and Infer sections of the config file. May be better in Global. This flexibility could be useful.

adpartin commented 3 weeks ago

As I understand, global parameters are expected to be available in all three model scripts. @jonesse3 Is that correct? I'm trying to remember the discussion we had against using global parameters. I think the reason was making sure that these global params are printed in --help when we call each one of the relevant scripts. For example, we want param model_file_name to be printed when we type python lgbm_train_improve.py --help and python lgbm_infer_improve.py --help @wilke @AlexandrNP do you remember?

wilke commented 3 weeks ago

Just a fun fact. You can set global parameters for all sections in the INI Config file. Right now the implementaion uses the DEFAULT section for this. The original request was to rename DEFAULT to GLOBAL. Everything which is assigned in DEFAULT is available in the Preprocess, Train and Infer config

adpartin commented 3 weeks ago

Good to know.

On Wed, Aug 21, 2024 at 6:47 PM Andreas Wilke @.***> wrote:

Just a fun fact. You can set global parameters for all sections in the INI Config file. Right now the implementaion uses the DEFAULT section for this. The original request was to rename DEFAULT to GLOBAL. Everything which is assigned in DEFAULT is available in the Preprocess, Train and Infer config

— Reply to this email directly, view it on GitHub https://github.com/JDACS4C-IMPROVE/IMPROVE/issues/55#issuecomment-2303305212, or unsubscribe https://github.com/notifications/unsubscribe-auth/AGFRLKH22OU6HTDAP6F7ZTLZSURJJAVCNFSM6AAAAABM4K27E6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMBTGMYDKMRRGI . You are receiving this because you were assigned.Message ID: @.***>

nkoussa commented 1 week ago

Note: If you use the [DEFAULT] section you have to have section headers (like [Preprocess], [Train], [Infer]) or it doesn't work. Like you can't just put things in [DEFAULT] and delete the [Infer] section. So I'm only documenting using the three sections, I'm not documenting the DEFAULT section since this seems prone to error. Might as well leave it as DEFAULT and close the issue, IMO.