New In This Release

1. Updates demo/app/tensor-stastistics.F90 to read and analyze a new ICAR training-data file format with merged image data, deleted halo data, and 50-time-step data groupings,
2. Similarly edits demo/app/train-cloud-microphysics.F90 and demo/train.sh (details below),
3. Makes several improvements to robustness in the demonstration app and fixes one example module.

Training Updates

With this release, putting the latest ICAR NetCDF training data files in a subdirectory demo/fiats-training-data and running something like

cd demo
./setup.sh
ln -s `find build -name train-cloud-microphysics`
./train.sh

should launch a run that does successively refined training of a cloud microphysics model by conditionally sampling the training data to flatten the output-tensor histograms.

This PR

1. Adjusts the demonstration application in demo/app/traini-cloud-microphysics.F90 and related files to process the ICAR training-data files that resulted from merging image-specific files into one regional file per 50 time steps;
2. Updates demo/train.sh to improve robustness by
   • Adding a check for the executable file and exiting if it's not found,
   • Adding default values for command-line arguments,
   • Removing unused command-line arguments, and
   • Updating to Julienne 3.6.1 to fix a crash that occurs when the network file already exists;
3. Adds file types to .gitignore;
4. Renames an example's supporting module to leverage default preprocessing;
5. Makes training output more verbose, including
   • more descriptions of intermediate steps and
   • printing epoch and cost function values.
6. Aggregates tensors across files for training.

Pull Request Log

• Update train-cloud-microphysics demo app for merged ICAR files by @rouson in #287
• Delete outdated/redundant/incomplete COPYRIGHT.txt file by @bonachea in #285

Full Changelog: 1.2.0...1.2.1