Merge pull request #350 from kaeldai/documentation/nwb_update

fixing issue with writing to nwb with mpi and adding some extra docum…

bmtk/utils/reports/spike_trains/spike_trains_api.py

@@ -24,7 +24,7 @@
 import warnings
 
 from .core import SortOrder
-from .spikes_file_writers import write_csv, write_sonata
+from .spikes_file_writers import write_csv, write_sonata, write_nwb
 
 
 class SpikeTrainsAPI(object):
@@ -169,7 +169,7 @@ def to_csv(self, path, mode='w', sort_order=SortOrder.none, **kwargs):
         write_csv(path=path, spiketrain_reader=self, mode=mode, sort_orders=sort_order, **kwargs)
 
     def to_nwb(self, path, mode='w', **kwargs):
-        raise NotImplemented()
+        write_nwb(path=path, spiketrain_reader=self, mode=mode, **kwargs)
 
     def merge(self, other):
         """Import Another SpikesTrain object into current file, always in-place.

bmtk/utils/reports/spike_trains/spikes_file_writers.py

@@ -24,7 +24,9 @@
 import csv
 import h5py
 import numpy as np
+from datetime import datetime
 
+import bmtk
 from .core import SortOrder, csv_headers, col_population, find_conversion
 from .core import MPI_rank, comm_barrier
 from bmtk.utils.sonata.utils import add_hdf5_magic, add_hdf5_version
@@ -145,3 +147,47 @@ def write_csv_itr(path, spiketrain_reader, mode='w', sort_order=SortOrder.none,
             csv_writer.writerow(c_data)
 
     comm_barrier()
+
+
+def write_nwb(path, spiketrain_reader, mode='w', include_population=True, units='ms', **kwargs):
+    import pynwb
+
+    path_dir = os.path.dirname(path)
+    if MPI_rank == 0 and path_dir and not os.path.exists(path_dir):
+        os.makedirs(path_dir)
+
+    if MPI_rank == 0:
+        # Last checked pynwb doesn't support writing on multiple cores, must let first core do all the
+        # writing to NWB.
+        nwbfile = pynwb.NWBFile(
+            session_description='BMTK {} generated NWB spikes file'.format(bmtk.__version__),
+            identifier='Generated in-silico, no session id',  # TODO: No idea what to put here?
+            session_start_time=datetime.now().astimezone(),
+            # experiment_description=str(session.experiment_metadata['experiment_id'])
+        )
+
+        if include_population:
+            nwbfile.add_unit_column(name="population", description="node population identifier")
+            add_unit = lambda nid, pop, st: nwbfile.add_unit(id=nid, spike_times=st, population=pop, node_id=nid)
+        else:
+            add_unit = lambda nid, pop, st: nwbfile.add_unit(id=nid, spike_times=st, node_id=nid)
+
+        nwbfile.add_unit_column(name="node_id", description="id of each node within a population")
+
+        for population in spiketrain_reader.populations:
+            for node_id in spiketrain_reader.node_ids(population=population):
+                spikes_times = spiketrain_reader.get_times(node_id=node_id, population=population)
+                if spikes_times is None or len(spikes_times) == 0:
+                    # No spikes for given node, don't try to write to nwb 
+                    continue
+
+                # sometimes bmtk/sonata may default to use different 32 or unsigned data-types which will cause
+                # nwb to throw a fit. Need to explicity convert data-types just in case.
+                spikes_times = spikes_times.astype('float64')
+                node_id = int(node_id)
+                add_unit(node_id, population, spikes_times)
+
+        with pynwb.NWBHDF5IO(path, mode) as io:
+            io.write(nwbfile)
+
+    comm_barrier()