Merge pull request #24 from francois-drielsma/develop

Add option to split output by input file, add shower track post-processors

spine/__init__.py

@@ -1,5 +1,3 @@
 """Top-level module of the SPICE source code."""
 
-from .data import *
-from .driver import Driver
 from .version import __version__

spine/driver.py

@@ -17,7 +17,6 @@
 
 import yaml
 import psutil
-import pathlib
 import numpy as np
 import torch
 
@@ -132,7 +131,8 @@ def __init__(self, cfg, rank=None):
             assert self.model is None or self.unwrap, (
                     "Must unwrap the model output to run analysis scripts.")
             self.watch.initialize('ana')
-            self.ana = AnaManager(ana, log_dir=self.log_dir, prefix=self.prefix)
+            self.ana = AnaManager(
+                    ana, log_dir=self.log_dir, prefix=self.log_prefix)
 
     def __len__(self):
         """Returns the number of events in the underlying reader object."""
@@ -232,7 +232,7 @@ def initialize_base(self, seed, dtype='float32', world_size=0,
                         log_dir='logs', prefix_log=False, overwrite_log=False,
                         parent_path=None, iterations=None, epochs=None,
                         unwrap=False, rank=None, log_step=1, distributed=False,
-                        train=None, verbosity='info'):
+                        split_output=False, train=None, verbosity='info'):
         """Initialize the base driver parameters.
 
         Parameters
@@ -263,6 +263,10 @@ def initialize_base(self, seed, dtype='float32', world_size=0,
             Number of iterations before the logging is called (1: every step)
         distributed : bool, default False
             If `True`, this process is distributed among multiple processes
+        train : dict, optional
+            Training configuration dictionary
+        split_output : bool, default False
+            Split the output of the process into one file per input file
         verbosity : int, default 'info'
             Verbosity level to pass to the `logging` module. Pick one of
             'debug', 'info', 'warning', 'error', 'critical'.
@@ -309,6 +313,7 @@ def initialize_base(self, seed, dtype='float32', world_size=0,
         self.unwrap = unwrap
         self.seed = seed
         self.log_step = log_step
+        self.split_output = split_output
 
         return train
 
@@ -358,15 +363,17 @@ def initialize_io(self, loader=None, reader=None, writer=None):
             self.iter_per_epoch = len(self.reader)
 
         # Fetch an appropriate common prefix for all input files
-        self.prefix = self.get_prefix(self.reader.file_paths)
+        self.log_prefix, self.output_prefix = self.get_prefixes(
+                self.reader.file_paths, self.split_output)
 
         # Initialize the data writer, if provided
         self.writer = None
         if writer is not None:
             assert self.loader is None or self.unwrap, (
                     "Must unwrap the model output to write it to file.")
             self.watch.initialize('write')
-            self.writer = writer_factory(writer, prefix=self.prefix)
+            self.writer = writer_factory(
+                    writer, prefix=self.output_prefix, split=self.split_output)
 
         # Harmonize the iterations and epochs parameters
         assert (self.iterations is None) or (self.epochs is None), (
@@ -379,17 +386,19 @@ def initialize_io(self, loader=None, reader=None, writer=None):
             self.iterations = self.epochs*self.iter_per_epoch
 
     @staticmethod
-    def get_prefix(file_paths):
+    def get_prefixes(file_paths, split_output):
         """Builds an appropriate output prefix based on the list of input files.
 
         Parameters
         ----------
         file_paths : List[str]
             List of input file paths
+        split_output : bool
+            Split the output of the process into one file per input file
 
         Returns
         -------
-        str
+        Union[str, List[str]]
             Shared input summary string to be used to prefix outputs
         """
         # Fetch file base names (ignore where they live)
@@ -409,8 +418,14 @@ def get_prefix(file_paths):
         last = last[0] if last[0] and last[0][0] != '.' else ''
 
         suffix = f'{first}--{len(file_names)-2}--{last}'
+        log_prefix = prefix + suffix
 
-        return prefix + suffix
+        # Always provide a single prefix for the log, adapt output prefix
+        if not split_output:
+            return log_prefix, log_prefix
+        else:
+            return (log_prefix,
+                    [os.path.splitext(name)[0] for name in file_names])
 
     def initialize_log(self):
         """Initialize the output log for this driver process."""
@@ -432,7 +447,7 @@ def initialize_log(self):
 
         # If requested, prefix the log name with the input file name
         if self.prefix_log:
-            log_name = f'{self.prefix}_{log_name}'
+            log_name = f'{self.log_prefix}_{log_name}'
 
         # Initialize the log
         log_path = os.path.join(self.log_dir, log_name)

spine/io/collate.py

@@ -6,7 +6,8 @@
 
 import numpy as np
 
-from spine import TensorBatch, IndexBatch, EdgeIndexBatch
+from spine.data import TensorBatch, IndexBatch, EdgeIndexBatch
+
 from spine.utils.geo import Geometry
 
 __all__ = ['CollateAll']

spine/io/factories.py

@@ -203,7 +203,7 @@ def reader_factory(reader_cfg):
     return instantiate(READER_DICT, reader_cfg)
 
 
-def writer_factory(writer_cfg, prefix=None):
+def writer_factory(writer_cfg, prefix=None, split=False):
     """Instantiates writer based on type specified in configuration under
     `io.writer.name`. The name must match the name of a class under
     `spine.io.writers`.
@@ -214,6 +214,8 @@ def writer_factory(writer_cfg, prefix=None):
         Writer configuration dictionary
     prefix : str, optional
         Input file prefix to use as an output name
+    split : bool, default False
+        Split the output into one file per input file
 
     Returns
     -------
@@ -225,4 +227,4 @@ def writer_factory(writer_cfg, prefix=None):
     Currently the choice is limited to `HDF5Writer` only.
     """
     # Initialize writer
-    return instantiate(WRITER_DICT, writer_cfg, prefix=prefix)
+    return instantiate(WRITER_DICT, writer_cfg, prefix=prefix, split=split)

spine/io/parse/cluster.py

@@ -10,7 +10,8 @@
 
 import numpy as np
 
-from spine import Meta
+from spine.data import Meta
+
 from spine.utils.globals import DELTA_SHP
 from spine.utils.particles import process_particle_event
 from spine.utils.ppn import image_coordinates

spine/io/parse/misc.py

@@ -10,7 +10,7 @@
 """
 
 
-from spine import Meta, RunInfo, Flash, CRTHit, Trigger, ObjectList
+from spine.data import Meta, RunInfo, Flash, CRTHit, Trigger, ObjectList
 
 from spine.utils.conditional import larcv
 

spine/io/parse/particle.py

@@ -12,7 +12,8 @@
 
 import numpy as np
 
-from spine import Meta, Particle, Neutrino, ObjectList
+from spine.data import Meta, Particle, Neutrino, ObjectList
+
 from spine.utils.globals import TRACK_SHP, PDG_TO_PID, PID_MASSES
 from spine.utils.particles import process_particles
 from spine.utils.ppn import get_ppn_labels, image_coordinates

spine/io/parse/sparse.py

@@ -9,7 +9,8 @@
 
 import numpy as np
 
-from spine import Meta
+from spine.data import Meta
+
 from spine.utils.globals import GHOST_SHP
 from spine.utils.ghost import compute_rescaled_charge
 from spine.utils.conditional import larcv

spine/io/write/hdf5.py

@@ -38,7 +38,8 @@ class HDF5Writer:
     name = 'hdf5'
 
     def __init__(self, file_name=None, keys=None, skip_keys=None,
-                 dummy_ds=None, overwrite=False, append=False, prefix=None):
+                 dummy_ds=None, overwrite=False, append=False,
+                 prefix=None, split=False):
         """Initializes the basics of the output file.
 
         Parameters
@@ -59,21 +60,38 @@ def __init__(self, file_name=None, keys=None, skip_keys=None,
         prefix : str, optional
             Input file prefix. It will be use to form the output file name,
             provided that no file_name is explicitely provided
+        split : bool, default False
+            If `True`, split the output to produce one file per input file
         """
-        # If the output file name is not provided, use the input file prefix
+        # If the output file name is not provided, use the input file prefix(es)
         if not file_name:
             assert prefix is not None, (
                     "If the output `file_name` is not provided, must provide"
                     "the input file `prefix` to build it from.")
-            file_name = prefix + '_spine.h5'
+            if not split:
+                file_name = f'{prefix}_spine.h5'
+            else:
+                file_name = [f'{pre}_spine.h5' for pre in prefix]
+
+        elif split:
+            file_name = [f'{file_name}_{i}' for i in range(len(prefix))]
 
-        # Check that output file does not already exist, if requestes
-        if not overwrite and os.path.isfile(file_name):
-            raise FileExistsError(f"File with name {file_name} already exists.")
+        # Check that the output file(s) do(es) not already exist, if requested
+        if not overwrite:
+            if not split:
+                if os.path.isfile(file_name):
+                    raise FileExistsError(
+                            f"File with name {file_name} already exists.")
+            else:
+                for f in file_name:
+                    if os.path.isfile(f):
+                        raise FileExistsError(
+                                f"File with name {f} already exists.")
 
         # Store persistent attributes
         self.file_name = file_name
         self.append = append
+        self.split = split
         self.ready = False
         self.object_dtypes = [] # TODO: make this a set
 
@@ -131,20 +149,22 @@ def create(self, data, cfg=None):
         for key in self.keys:
             self.register_key(data, key)
 
-        # Initialize the output HDF5 file
-        with h5py.File(self.file_name, 'w') as out_file:
-            # Initialize the info dataset that stores environment parameters
-            if cfg is not None:
-                out_file.create_dataset(
-                        'info', (0,), maxshape=(None,), dtype=None)
-                out_file['info'].attrs['cfg'] = yaml.dump(cfg)
-                out_file['info'].attrs['version'] = __version__
+        # Initialize the output HDF5 file(s)
+        file_names = [self.file_name] if not self.split else self.file_name
+        for file_name in file_names:
+            with h5py.File(file_name, 'w') as out_file:
+                # Initialize the info dataset that stores environment parameters
+                if cfg is not None:
+                    out_file.create_dataset(
+                            'info', (0,), maxshape=(None,), dtype=None)
+                    out_file['info'].attrs['cfg'] = yaml.dump(cfg)
+                    out_file['info'].attrs['version'] = __version__
 
-            # Initialize the event dataset and their reference array datasets
-            self.initialize_datasets(out_file)
+                # Initialize the event dataset and their reference array datasets
+                self.initialize_datasets(out_file)
 
-            # Mark file as ready for use
-            self.ready = True
+        # Mark file(s) as ready for use
+        self.ready = True
 
     def get_stored_keys(self, data):
         """Get the list of data product keys to store.
@@ -432,23 +452,45 @@ def __call__(self, data, cfg=None):
             self.create(data, cfg)
             self.ready = True
 
-        # Append file
-        with h5py.File(self.file_name, 'a') as out_file:
-            # Loop over batch IDs
-            for batch_id in range(batch_size):
-                # Initialize a new event
-                event = np.empty(1, self.event_dtype)
-
-                # Initialize a dictionary of references to be passed to the
-                # event dataset and store the input and result keys
-                for key in self.keys:
-                    self.append_key(out_file, event, data, key, batch_id)
-
-                # Append event
-                event_id = len(out_file['events'])
-                event_ds = out_file['events']
-                event_ds.resize(event_id + 1, axis=0) # pylint: disable=E1101
-                event_ds[event_id] = event
+        # Append file(s)
+        if not self.split:
+            with h5py.File(self.file_name, 'a') as out_file:
+                # Loop over batch IDs
+                for batch_id in range(batch_size):
+                    self.append_entry(out_file, data, batch_id)
+
+        else:
+            file_ids = data['file_index']
+            for file_id in np.unique(file_ids):
+                with h5py.File(self.file_name[file_id], 'a') as out_file:
+                    for batch_id in np.where(file_ids == file_id)[0]:
+                        self.append_entry(out_file, data, batch_id)
+
+    def append_entry(self, out_file, data, batch_id):
+        """Stores one entry.
+
+        Parameters
+        ----------
+        out_file : h5py.File
+            HDF5 file instance
+        data : dict
+            Dictionary of data products
+        batch_id : int
+            Batch ID to be stored
+        """
+        # Initialize a new event
+        event = np.empty(1, self.event_dtype)
+
+        # Initialize a dictionary of references to be passed to the
+        # event dataset and store the input and result keys
+        for key in self.keys:
+            self.append_key(out_file, event, data, key, batch_id)
+
+        # Append event
+        event_id = len(out_file['events'])
+        event_ds = out_file['events']
+        event_ds.resize(event_id + 1, axis=0) # pylint: disable=E1101
+        event_ds[event_id] = event
 
     def append_key(self, out_file, event, data, key, batch_id):
         """Stores data key in a specific dataset of an HDF5 file.

spine/model/grappa.py

@@ -8,7 +8,8 @@
 from .layer.factories import final_factory
 from .layer.gnn.factories import *
 
-from spine import TensorBatch, IndexBatch, EdgeIndexBatch
+from spine.data import TensorBatch, IndexBatch, EdgeIndexBatch
+
 from spine.utils.globals import (
         BATCH_COL, COORD_COLS, CLUST_COL, GROUP_COL, SHAPE_COL, LOWES_SHP)
 from spine.utils.enums import enum_factory

spine/model/layer/common/dbscan.py

@@ -4,7 +4,8 @@
 import torch
 from sklearn.cluster import DBSCAN as sklearn_dbscan
 
-from spine import TensorBatch, IndexBatch
+from spine.data import TensorBatch, IndexBatch
+
 from spine.utils.globals import (
         SHOWR_SHP, TRACK_SHP, MICHL_SHP, DELTA_SHP, COORD_COLS, PPN_SHAPE_COL,
         COORD_START_COLS, COORD_END_COLS)

spine/model/layer/common/final.py

@@ -4,7 +4,7 @@
 
 from torch import nn
 
-from spine import TensorBatch
+from spine.data import TensorBatch
 
 from .mlp import MLP
 from .evidential import EvidentialModel

spine/model/layer/gnn/encode/cnn.py

@@ -4,7 +4,8 @@
 
 from spine.model.layer.cnn.encoder import SparseResidualEncoder
 
-from spine import TensorBatch, IndexBatch
+from spine.data import TensorBatch, IndexBatch
+
 from spine.utils.globals import BATCH_COL
 
 __all__ = ['ClustCNNNodeEncoder', 'ClustCNNEdgeEncoder', 'ClustCNNGlobalEncoder']

spine/model/layer/gnn/encode/geometric.py

@@ -1,7 +1,8 @@
 import numpy as np
 import torch
 
-from spine import TensorBatch
+from spine.data import TensorBatch
+
 from spine.utils.torch_local import local_cdist
 from spine.utils.globals import COORD_COLS, VALUE_COL, SHAPE_COL
 from spine.utils.gnn.cluster import (