Closed tomasvanoyen closed 1 year ago
Hi @tomasvanoyen, glad to hear you got some batches made.
Those config files are old experiments we ran. We chose not to include them in the library so as not to clog it up with lots of unnecessary configs. However, those old experiments informed the model architecture defined in multimodel.yaml
. So this one can be seen as our best model so far. I'd recommend you start from there. I'm sure there are more improvements to be made.
Hi @dfulu,
thanks for clarifying on the model configuration. I will keep you posted on possible forthcoming successes and failures. regards, Tomas
Trying to run scripts as described @dfulu from
experiments.sh
. Saving validation and training batches locally is okay.Yet, confused for the training part. it appears that several of the configuration files as indicated are not present in the code base; e.g. reference is given to
conv3d_sat_nwp_v4.yaml
andconv3d_sat_nwp_v6.yaml
which appears missing in themodel
folder. Could you describe the changes w.r.t.conv3d_sat_nwp.yaml
with other model configuration files such asmultimodel.yaml
or just add it to the repo?Thanks!
Tomas