1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
|
import h5py
import numpy as np
from packaging import version
from .exceptions import MultiRunError
class FileWriter:
"""Write data in European XFEL HDF5 format
This is intended to allow copying a subset of data into a smaller,
more portable file.
"""
def __init__(self, path, data):
self.file = h5py.File(path, 'w')
self.data = data
self.indexes = {} # {path: (first, count)}
self.data_sources = set()
def prepare_source(self, source):
"""Prepare all the datasets for one source.
We do this as a separate step so the contents of the file are defined
together before the main data.
"""
for key in sorted(self.data.keys_for_source(source)):
path = f"{self._section(source)}/{source}/{key.replace('.', '/')}"
nentries = self._guess_number_of_storing_entries(source, key)
src_ds1 = self.data[source].files[0].file[path]
self.file.create_dataset_like(
path, src_ds1, shape=(nentries,) + src_ds1.shape[1:],
# Corrected detector data has maxshape==shape, but if any max
# dim is smaller than the chunk size, h5py complains. Making
# the first dimension unlimited avoids this.
maxshape=(None,) + src_ds1.shape[1:],
)
if source in self.data.instrument_sources:
self.data_sources.add(f"INSTRUMENT/{source}/{key.partition('.')[0]}")
if source not in self.data.instrument_sources:
self.data_sources.add(f"CONTROL/{source}")
def _guess_number_of_storing_entries(self, source, key):
"""Provide the length for the initial dataset to create.
May be overridden in subclasses.
"""
return self.data.get_data_counts(source, key).sum()
def _section(self, source):
if source in self.data.instrument_sources:
return 'INSTRUMENT'
else:
return 'CONTROL'
def copy_dataset(self, source, key):
"""Copy data into a dataset"""
a = self.data.get_array(source, key)
path = f"{self._section(source)}/{source}/{key.replace('.', '/')}"
self.file[path][:] = a.values
self._make_index(source, key, a.coords['trainId'].values)
def _make_index(self, source, key, data_tids):
# Original files contain exactly 1 entry per train for control data,
# but if one file starts before another, there can be some values
# missing when we collect several files together. We don't try to
# extrapolate to fill missing data, so some counts may be 0.
if source in self.data.instrument_sources:
index_path = source + '/' + key.partition('.')[0]
else:
index_path = source
if index_path not in self.indexes:
if source not in self.data.instrument_sources:
assert len(np.unique(data_tids)) == len(data_tids),\
"Duplicate train IDs in control data!"
self.indexes[index_path] = self._generate_index(data_tids)
def _generate_index(self, data_tids):
"""Convert an array of train IDs to first/count for each train"""
assert (np.diff(data_tids) >= 0).all(), "Out-of-order train IDs"
counts = np.array([np.count_nonzero(t == data_tids)
for t in self.data.train_ids], dtype=np.uint64)
firsts = np.zeros_like(counts)
firsts[1:] = np.cumsum(counts)[:-1] # firsts[0] is always 0
return firsts, counts
def copy_source(self, source):
"""Copy data for all keys of one source"""
for key in self.data.keys_for_source(source):
self.copy_dataset(source, key)
def write_train_ids(self):
self.file.create_dataset(
'INDEX/trainId', data=self.data.train_ids, dtype='u8'
)
train_timestamps = self.data.train_timestamps()
if not np.all(np.isnat(train_timestamps)):
self.file.create_dataset(
'INDEX/timestamp', data=train_timestamps.astype(np.uint64)
)
def write_indexes(self):
"""Write the INDEX information for all data we've copied"""
for groupname, (first, count) in self.indexes.items():
group = self.file.create_group(f'INDEX/{groupname}')
group.create_dataset('first', data=first, dtype=np.uint64)
group.create_dataset('count', data=count, dtype=np.uint64)
def write_metadata(self):
try:
metadata = self.data.run_metadata()
except MultiRunError:
metadata = {}
metadata_grp = self.file.create_group('METADATA')
format_version = version.parse(metadata.get('dataFormatVersion'))
if format_version >= version.parse("1.0"):
# We don't care about the differences between version 1.0/1.1/1.2,
# so for simplicity we stick to the 1.0 format.
metadata["dataFormatVersion"] = "1.0"
self.write_sources(metadata_grp.create_group('dataSources'))
# File format 1.0 should also have INDEX/flag
self.file.create_dataset('INDEX/flag', data=self.gather_flag())
for key, val in metadata.items():
metadata_grp[key] = [val]
else:
# File format '0.5': source lists directly in METADATA
self.write_sources(metadata_grp)
def write_sources(self, data_sources_grp: h5py.Group):
"""Write the METADATA section, including lists of sources"""
vlen_bytes = h5py.special_dtype(vlen=bytes)
data_sources = sorted(self.data_sources)
N = len(data_sources)
sources_ds = data_sources_grp.create_dataset(
'dataSourceId', (N,), dtype=vlen_bytes, maxshape=(None,)
)
sources_ds[:] = data_sources
root_ds = data_sources_grp.create_dataset(
'root', (N,), dtype=vlen_bytes, maxshape=(None,)
)
root_ds[:] = [ds.split('/', 1)[0] for ds in data_sources]
devices_ds = data_sources_grp.create_dataset(
'deviceId', (N,), dtype=vlen_bytes, maxshape=(None,)
)
devices_ds[:] = [ds.split('/', 1)[1] for ds in data_sources]
def gather_flag(self):
"""Make the array for INDEX/flag.
Trains are valid (1) if they are valid in *any* of the source files.
"""
tid_arr = np.asarray(self.data.train_ids, dtype=np.uint64)
flag = np.zeros_like(tid_arr, dtype=np.int32)
for fa in self.data.files:
mask_valid = np.isin(tid_arr, fa.valid_train_ids)
flag[mask_valid] = 1
return flag
def set_writer(self):
"""Record the package & version writing the file in an attribute"""
from . import __version__
self.file.attrs['writer'] = 'extra_data {}'.format(__version__)
def write(self):
d = self.data
self.set_writer()
self.write_train_ids()
for source in d.all_sources:
self.prepare_source(source)
self.write_metadata()
for source in d.all_sources:
self.copy_source(source)
self.write_indexes()
class VirtualFileWriter(FileWriter):
"""Write virtual datasets in European XFEL format
The new files refer to the original data files, so they aren't portable,
but they provide more convenient access by reassembling data spread over
several sequence files.
"""
def __init__(self, path, data):
if not hasattr(h5py, 'VirtualLayout'):
raise Exception("Creating virtual datasets requires HDF5 1.10 "
"and h5py 2.9")
super().__init__(path, data)
def _assemble_data(self, keydata):
"""Assemble chunks of data into a virtual layout"""
# Create the layout, which will describe what data is where
layout = h5py.VirtualLayout(shape=keydata.shape, dtype=keydata.dtype)
# Map each chunk into the relevant part of the layout
output_cursor = 0
for chunk in keydata._data_chunks_nonempty:
n = chunk.total_count
src = h5py.VirtualSource(chunk.dataset)
layout[output_cursor : output_cursor + n] = src[chunk.slice]
output_cursor += n
assert output_cursor == layout.shape[0]
return layout
# In big detector data, these fields are like extra indexes.
# So we'll copy them to the output file for fast access, rather than
# making virtual datasets.
copy_keys = {'image.pulseId', 'image.cellId'}
def prepare_source(self, source):
srcdata = self.data[source]
grp_out = self.file.require_group(f'{srcdata.section}/{source}')
grp_out.attrs['source_files'] = sorted([f.filename for f in srcdata.files])
for key in srcdata.keys():
if key in self.copy_keys:
self.copy_dataset(source, key)
else:
self.add_dataset(source, key)
# Add a link in RUN for control sources
if srcdata.is_control:
src_file = srcdata.files[0]
run_path = f'RUN/{source}'
self.file[run_path] = h5py.ExternalLink(src_file.filename, run_path)
def copy_dataset(self, source, key):
"""Copy data as a new dataset"""
a = self.data.get_array(source, key)
path = f"{self._section(source)}/{source}/{key.replace('.', '/')}"
self.file.create_dataset(path, data=a.values, compression='gzip')
self._make_index(source, key, a.coords['trainId'].values)
def add_dataset(self, source, key):
keydata = self.data[source, key]
if keydata.shape[0] == 0: # No data
# Make the dataset virtual even with no source data to map.
# This workaround will hopefully become unnecessary from h5py 3.14
parent_path, name = keydata.hdf5_data_path.rsplit('/', 1)
group = self.file.require_group(parent_path)
dcpl = h5py.h5p.create(h5py.h5p.DATASET_CREATE)
dcpl.set_layout(h5py.h5d.VIRTUAL)
h5py.h5d.create(
group.id,
name=name.encode(),
tid=h5py.h5t.py_create(keydata.dtype, logical=1),
space=h5py.h5s.create_simple(keydata.shape),
dcpl=dcpl
)
else:
layout = self._assemble_data(keydata)
self.file.create_virtual_dataset(keydata.hdf5_data_path, layout)
self._make_index(source, key, keydata.train_id_coordinates())
if source in self.data.instrument_sources:
self.data_sources.add(f"INSTRUMENT/{source}/{key.partition('.')[0]}")
else:
self.data_sources.add(f"CONTROL/{source}")
return keydata.hdf5_data_path
def copy_source(self, source):
pass # Override base class copying data
|
