Test QuickStart

Step 3 of the Quickstart reads:

Lines 81 to 83 in f94b862

    
           **Step 3** 
        
           Train your model on the training data and validate using the validation data. If you wish to use something like a CNN, you will probably want to separate the variables into channels and broadcast scalars into vectors of the same dimension as vertically-resolved variables. Methods to do this can be found in the [climsim_utils/data_utils.py](https://github.com/leap-stc/ClimSim/blob/main/climsim_utils/data_utils.py) script.

Is it reasonable to expect that our target audience will know how to do this? As someone who has never trained an ML model before, I personally don't know where to begin, but I don't think I'm the target audience.

(Alternatively, if it were to be catered to a more beginner audience, the Quickstart Step 3 could simply describe how to replicate the paper results by running the baseline_models/ (or one of them) against the subsampled training data?)

If the expectation is that the target Quickstart audience knows how to train a model against this data without further instruction, then I'll just have to recuse myself from being a tester for this section until I know how to do that (which probably won't be today)! 😄

I agree. We cannot expect our audience to know how to train a model on this new dataset. I train ML models on other datasets, but the dataset really makes a big difference to the workflow. I consider myself as the target audience, but I would not know how to quickly train such a model. (In fact, I would expect that this would take days to weeks if I had to do it from scratch.)

(Alternatively, if it were to be catered to a more beginner audience, the Quickstart Step 3 could simply describe how to replicate the paper results by running the baseline_models/ (or one of them) against the subsampled training data?)

Yes, this is a good suggestion.

I think the quickstart is in a condition ready for testing.

@jerrylin96 did you close this issue on purpose? I have not completed the testing of the quickstart. In fact, I was waiting on instructions, see #55 (comment).

	Step 3

	Train your model on the training data and validate using the validation data. If you wish to use something like a CNN, you will probably want to separate the variables into channels and broadcast scalars into vectors of the same dimension as vertically-resolved variables. Methods to do this can be found in the [climsim_utils/data_utils.py](https://github.com/leap-stc/ClimSim/blob/main/climsim_utils/data_utils.py) script.