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
287
288
289
290
291
292
|
# -*- coding: utf-8 -*-
# Copyright © 2007-2008 Stockholm TreeAligner Project
# Author: Torsten Marek <shlomme@gmx.net>
# Licensed under the GNU GPLv2
from collections import defaultdict
from itertools import count
import logging
def get_version_string():
import nltk
return "nltk %s" % (nltk.__version__, )
from nltk_contrib.tiger.index import IndexNodeId, CONTINUOUS, DISCONTINUOUS, gorn2db
from nltk_contrib.tiger.graph import NodeType, veeroot_graph, DEFAULT_VROOT_EDGE_LABEL
__all__ = ("TigerCorpusIndexer",)
# think about splitting feature_values table into several small tables.
# add NOT NULLS everywhere
# TODO: create proper progress reporter interface, hand in
INDEX_VERSION = 3
class _Tables(object):
FEATURES = """CREATE TABLE features
(id INTEGER PRIMARY KEY, order_id INTEGER, name TEXT, domain TEXT)"""
FEATURE_VALUES = """CREATE TABLE feature_values
(feature_id INTEGER, value_id INTEGER, value TEXT, description TEXT,
UNIQUE(feature_id, value_id))"""
EDGE_LABELS = """CREATE TABLE edge_labels
(id INTEGER PRIMARY KEY, label TEXT UNIQUE, description TEXT)"""
SECEDGE_LABELS = """CREATE TABLE secedge_labels
(id INTEGER PRIMARY KEY, label TEXT UNIQUE, description TEXT)"""
GRAPHS = """CREATE TABLE graphs
(id INTEGER PRIMARY KEY, xml_graph_id TEXT, data BLOB)"""
METADATA = """CREATE TABLE corpus_metadata
(key TEXT UNIQUE, value TEXT)"""
INDEX_METADATA = """CREATE TABLE index_metadata
(key TEXT UNIQUE, value NONE)"""
NODE_DATA = """CREATE TABLE node_data
(id INTEGER PRIMARY KEY, xml_node_id TEXT, edge_label INTEGER,
gorn_address BLOB, continuity INTEGER, arity INTEGER,
tokenarity INTEGER, left_corner INTEGER, right_corner INTEGER, token_order INTEGER)"""
FEATURE_IIDX_TEMPLATE = """CREATE TABLE feature_iidx_%s
(node_id INTEGER PRIMARY KEY NOT NULL, value_id INTEGER NOT NULL)"""
SECEDGES = """CREATE TABLE secedges
(origin_id INT, target_id INT, label_id INT)"""
class TigerCorpusIndexer(object):
def __init__(self, db, graph_serializer, progress = False, always_veeroot = True):
self._db = db
self._cursor = db.cursor()
self._progress = progress
self._always_veeroot = always_veeroot
self._graphs = 0
self._cursor.execute(_Tables.FEATURES)
self._cursor.execute(_Tables.FEATURE_VALUES)
self._cursor.execute(_Tables.EDGE_LABELS)
self._cursor.execute(_Tables.SECEDGE_LABELS)
self._cursor.execute(_Tables.GRAPHS)
self._cursor.execute(_Tables.METADATA)
self._cursor.execute(_Tables.INDEX_METADATA)
self._cursor.execute(_Tables.NODE_DATA)
self._cursor.execute(_Tables.SECEDGES)
self._serializer = graph_serializer
self._open_list_features = []
self._feature_count = {
NodeType.TERMINAL: 0,
NodeType.NONTERMINAL: 0}
self._feature_iidx_stmts = {}
self._feature_value_maps = {}
self._insert_lists = defaultdict(list)
self._store_creator_metadata()
def _store_creator_metadata(self):
self._add_index_metadata(creator=get_version_string(), index_version=INDEX_VERSION)
def _add_index_metadata(self, **kwargs):
self._cursor.executemany("INSERT INTO index_metadata (key, value) VALUES (?, ?)",
kwargs.iteritems())
def set_metadata(self, metadata):
self._cursor.executemany("INSERT INTO corpus_metadata (key, value) VALUES (?, ?)",
metadata.iteritems())
def add_feature(self, feature_name, domain, feature_values):
order_id = self._feature_count[domain]
self._feature_count[domain] += 1
self._cursor.execute("INSERT INTO features (order_id, name, domain) VALUES (?, ?, ?)",
(order_id, feature_name, domain.key))
feature_id = self._cursor.lastrowid
if len(feature_values) > 0:
value_map = dict((feature_value, idx) for idx, feature_value in enumerate(feature_values))
self._cursor.executemany("INSERT INTO feature_values (feature_id, value_id, value, description) VALUES (?, ?, ?, ?)",
((feature_id, value_map[value], value, description)
for value, description in feature_values.iteritems()))
else:
value_map = defaultdict(count().next)
self._open_list_features.append((feature_id, value_map))
self._feature_value_maps[feature_name] = (value_map, domain)
self._serializer.add_feature_value_map(feature_name, domain, order_id, value_map)
self._create_feature_value_index(feature_name)
return feature_id
def set_edge_labels(self, edge_labels):
if self._always_veeroot:
assert DEFAULT_VROOT_EDGE_LABEL in edge_labels, "no neutral edge label"
self._cursor.executemany("INSERT INTO edge_labels (id, label, description) VALUES (?, ?, ?)",
((idx, e[0], e[1]) for idx, e in enumerate(edge_labels.iteritems())))
self._edge_label_map = dict(self._cursor.execute("SELECT label, id FROM edge_labels"))
self._serializer.set_edge_label_map(self._edge_label_map)
def set_secedge_labels(self, secedge_labels):
self._cursor.executemany("INSERT INTO secedge_labels (id, label, description) VALUES (?, ?, ?)",
((idx, e[0], e[1]) for idx, e in enumerate(secedge_labels.iteritems())))
self._secedge_label_map = dict(self._cursor.execute("SELECT label, id FROM secedge_labels"))
self._serializer.set_secedge_label_map(self._secedge_label_map)
def _create_feature_value_index(self, feature_name):
feature_name = str(feature_name)
assert feature_name.isalpha()
self._cursor.execute(_Tables.FEATURE_IIDX_TEMPLATE % (feature_name,))
self._feature_iidx_stmts[feature_name] = "INSERT INTO feature_iidx_%s (node_id, value_id) VALUES (?, ?)" % (feature_name,)
def get_terminal_index_data(self, node, node_ids):
return (node_ids[node.id].to_int(), node.id, self._edge_label_map.get(node.edge_label, None),
gorn2db(node.gorn_address), node.order)
def get_nonterminal_index_data(self, node, node_ids):
return (node_ids[node.id].to_int(), node.id, self._edge_label_map.get(node.edge_label, None),
gorn2db(node.gorn_address),
(CONTINUOUS if node.is_continuous else DISCONTINUOUS),
node.arity, node.token_arity,
node_ids[node.left_corner].to_int(), node_ids[node.right_corner].to_int(),
node.children_type)
def _store_node_data(self, graph, node_ids):
nonterminals, terminals = graph.compute_node_information()
self._cursor.executemany("""INSERT INTO node_data
(id, xml_node_id, edge_label, gorn_address, continuity, arity, tokenarity, left_corner, right_corner, token_order)
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)""",
(self.get_nonterminal_index_data(nt, node_ids) for nt in nonterminals))
self._cursor.executemany("INSERT INTO node_data (id, xml_node_id, edge_label, gorn_address, token_order, continuity) VALUES (?, ?, ?, ?, ?, 0)",
(self.get_terminal_index_data(t, node_ids) for t in terminals))
def _index_feature_values(self, graph, node_ids):
for node in graph:
for feature_name, feature_value in node.features.iteritems():
value_map, domain = self._feature_value_maps[feature_name]
assert node.TYPE is domain
self._insert_lists[feature_name].append((node_ids[node.id].to_int(), value_map[feature_value]))
if self._graphs % 1000 == 0:
self._flush_node_feature_values()
def _index_secedges(self, graph, node_ids):
for node in graph:
if node.secedges is not None:
self._cursor.executemany(
"INSERT INTO secedges (origin_id, target_id, label_id) VALUES (?, ?, ?)",
((node_ids[node.id].to_int(), node_ids[graph.nodes[target_node].id].to_int(),
self._secedge_label_map[label])
for label, target_node in node.secedges))
def _flush_node_feature_values(self):
for feature_name, values in self._insert_lists.iteritems():
self._cursor.executemany(self._feature_iidx_stmts[feature_name], values)
self._insert_lists = defaultdict(list)
def _convert_ids(self, graph, node_ids): # split out into separate method
def _convert_edgelist(l):
return [(label, node_ids[target_xml_id])
for label, target_xml_id in l]
graph.id = self._graphs
graph.root_id = node_ids[graph.root_id]
for xml_node_id in graph.nodes.keys():
node = graph.nodes.pop(xml_node_id)
node.id = node_ids[node.id]
graph.nodes[node_ids[xml_node_id]] = node
if node.secedges:
node.secedges = _convert_edgelist(node.secedges)
if node.TYPE is NodeType.NONTERMINAL:
node.edges = _convert_edgelist(node.edges)
def add_graph(self, graph):
try:
roots = graph.get_roots()
except KeyError, e:
logging.error("Graph %s is faulty: node %s referenced more than once.",
graph.id, e.args[0])
return
if self._always_veeroot:
veeroot_graph(graph, roots)
else:
assert len(roots) == 1, "No auto-veerooting, but several unconnected subgraphs %s in %s." % (roots, graph.id)
node_ids = dict((xml_node_id, IndexNodeId(self._graphs, idx))
for idx, xml_node_id in enumerate(graph.nodes))
xml_id = graph.id
self._store_node_data(graph, node_ids)
self._index_feature_values(graph, node_ids)
self._index_secedges(graph, node_ids)
self._convert_ids(graph, node_ids)
self._cursor.execute("INSERT INTO graphs (id, xml_graph_id, data) VALUES (?, ?, ?)",
(self._graphs, xml_id, buffer(self._serializer.serialize_graph(graph))))
self._graphs += 1
if self._progress and self._graphs % 100 == 0:
print self._graphs
def finalize(self, optimize = True):
if self._progress:
print "finalize"
self._flush_node_feature_values()
if self._progress:
print "inserting feature values"
for feature_id, feature_value_map in self._open_list_features:
self._cursor.executemany("INSERT INTO feature_values (feature_id, value_id, value) VALUES (?, ?, ?)",
((feature_id, value_id, value)
for value, value_id in feature_value_map.iteritems()))
del self._open_list_features
if self._progress:
print "Committing database"
self._db.commit()
self._cursor.execute("CREATE INDEX feature_id_idx ON feature_values (feature_id)")
for feature_name in self._feature_value_maps:
if self._progress:
print "creating index for feature '%s'" % (feature_name,)
self._cursor.execute("CREATE INDEX %s_iidx_idx ON feature_iidx_%s (value_id)" % (feature_name, feature_name))
if self._progress:
print "creating index for xml node ids"
self._cursor.execute("CREATE UNIQUE INDEX xml_node_id_idx ON node_data (xml_node_id)")
if self._progress:
print "creating index for xml graph ids"
self._cursor.execute("CREATE UNIQUE INDEX xml_graph_id_idx ON graphs (xml_graph_id)")
if self._progress:
print "creating secedge indices"
self._cursor.execute("CREATE INDEX se_origin_idx ON secedges (origin_id)")
self._cursor.execute("CREATE INDEX se_target_idx ON secedges (target_id)")
self._db.commit()
if optimize:
if self._progress:
print "Optimizing database"
self._db.execute("VACUUM")
self._add_index_metadata(finished = True)
self._db.commit()
self._db = None
self._cursor = None
|
