Add a bash file for training and creating experiment

[xushanthu-2014]

Maybe we can add a bash file for training and creating experiment which will explicitly avoid:

1. submit jobs 3 times because slurm can only handle 1000 tasks per time;
2. mkdir for new experiment and copy the .py and .slurm file into new directory.
   maybe we can also generate:
3. a list of number of gpis if best model or shap values are not saved...because there is always bug where shape of shap values are not equal to shape of lat/lon/gpi/trained_model....
4. a list of RMSE and MAE or so instead of .log file. If we submit 3 times, .log files of same job_id will be overwritten.

[rogerkuou]

fixed in #46