PompousParrot44 the fundamental difference is that artifacts are uploaded manually (i.e. a user will specifically "ask" to upload an artifact), models are logged automatically and a user might not want them uploaded (imagine debugging sessions, or testing).
By adding the 'upload_uri' arguments, you can specify to trains that you want all models to be automatically uploaded (not just logged).
Now here is the nice thing, when running using the trains-agent, you can have:
Always upload the model by configuring the https://github.com/allegroai/trains-agent/blob/699d13bbb34649c7e5337b4187cda59b7fa6fd33/docs/trains.conf#L262 In the Web UI under the execution tab, set the "output destination". It is equivalent to setting the output_uri
so i was expecting that uploaded model will be for example LightGBM.1104445eca4749f89962669200481397/models/Model%20object/model.pkl
so just under models dir rather than artifact... any way to achieve this or i should just treat it as artifact ?
PompousParrot44
you can always manually store/load models, example: https://github.com/allegroai/trains/blob/65a4aa7aa90fc867993cf0d5e36c214e6c044270/examples/reporting/model_config.py#L35 Sure, you can patch any frame work with something similar to what we do in xgboost, any such PR will be greatly appreciated! https://github.com/allegroai/trains/blob/master/trains/binding/frameworks/xgboost_bind.py
AgitatedDove14 Morning... so what should the value of "upload_uri" to set to, fileserver_url e.g. http://localhost:8081 ?
AgitatedDove14 when using OutputModel(task, name='LightGBM model', framework='LightGBM').update_weights(f"{args.out}/model.pkl") i am seeing this in the logs No output storage destination defined, registering local model /tmp/model.pkl when i got to trains UI.. i see the model name and details but when i try to download it point to the path file:///tmp/model.pkl which is incorrect wondering how to fix it
looking at the code https://github.com/allegroai/trains/blob/65a4aa7aa90fc867993cf0d5e36c214e6c044270/trains/model.py#L1146 this happens when storage_uri is not defined where as i have this under trains.conf so task should have it ?
seems like setting to fileserver_url did the trick
thanks AgitatedDove14 for the links.. seems like i might try first one if it works out .. before going the route to create a full framework support as in our case team uses multiple different frameworks
AgitatedDove14 it seems uploading artifact and uploading models are two different things when it comes to treating fileserver... as when i upload artifact it works as expected but when uploading model using outputmodel class, it wants output_uri path.. wondering how can i as it to store it under the fileserver like artifacts LightGBM.1104445eca4749f89962669200481397/artifacts/Model%20object/model.pkl
so what should the value of "upload_uri" to set to,
fileserver_url
e.g.
?
yes, that would work.