MLMPR

To process deep learning codes using TensorFlow on the Wilson Cluster, see the DLRunScripts package. For legacy Theano and Caffe, see the scripts here, but note that they have not been updated for the new batch processing system on the Wilson Cluster.

• Caffe/ - scripts and prototxts for running vertex finding.
• mnvtf/ - TensorFlow code for the MINERvA nuclear targets vertex finder and some legacy run scripts.
  • data_constants.py - strings, structures, and constants used to specify HDF5 and TFRecord files.
  • data_readers.py - specialized classes for reading TFRecord files using the "old" batch-queue API.
  • dset_data_readers.py - specialized classes for reading TFRecord files using the tf.data.Dataset API.
  • evtid_utils.py - utilties for decoding the eventid fields in 64 bit or double-32 bit combos.
  • hf5_readers.py - specialized classes for reading HDF5 files.
  • models_menagerie.py - functions for specifying models (as dictionaries to be parsed by the mnvtf mini-framework).
  • models_tricolumnar.py - specialized classes for parsing models specified as dictionaries into three-branch convolutional models.
  • reader_sqlite.py - specialized classes for reading predictions recorded as SQLite files (requires sqlalchemy).
  • reader_text.py - specialized classes for reading predictions recorded as text files.
  • recorder_sqlite.py - specialized classes for recording predictions as SQLite files (requires sqlalchemy).
  • recorder_text.py - specialized classes for recording predictions as text files.
  • runners.py - specialized classes for running training and inference tasks.
  • utils.py - misc. utility functions.
• archive/ - old code kept visible for reference.
• dset_visualize.py - event display viewer that may consume HDF5 or TFRecord files.
• examine_hdf5.py - simple script to examine the structure and sizes of a MINERvA HDF5 file.
• hdf5_to_tfrec_minerva_xtxutuvtv.py - script for converting HDF5 files to TensorFlow TFRecord.
• horovod_mnist.py - script from Uber to run MNIST classification using Horovod.
• horovod_test.py - test to see if we can initialize Horovod.
• mnv_run_st_epsilon.py - run classification using the "space-time" version of the "epsilon" network architecture for vertex finding in the target analysis.
• plane_codes.py - legacy utilities code for converting the 'old' MINERvA framework plane id numbers into sequential planecodes.
• tfrec_examiner.py - script that checks the number of records in a TFRecord file and prints the eventid values to a log.
• txt_to_sqlite.py - converter script for writing text-based prediction files into SQLite files.