mirror of
https://github.com/adlerosn/rede-especificacoes-tecnicas-em-redes
synced 2024-07-05 17:00:12 +00:00
149 lines
5.7 KiB
Python
Executable File
149 lines
5.7 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
# -*- encoding: utf-8 -*-
|
|
|
|
import json
|
|
import sqlite3
|
|
import graphviz
|
|
from pathlib import Path
|
|
|
|
from .documents import fromFile as DocumentFromFile
|
|
from .document_finder import classes as docClasses
|
|
from .document_finder import find_references as referenceFinder
|
|
|
|
|
|
def generate_graph(rootdoc='rootdoc.txt', grapfn='graph.json', keep_temporal_context=True):
|
|
rootsrc, rootname = Path(rootdoc).read_text().splitlines()
|
|
analyzedDocPaths = list()
|
|
pendingDocCchMgr = [docClasses[rootsrc](rootname)]
|
|
graph = dict()
|
|
while len(pendingDocCchMgr) > 0:
|
|
docCchMgr, *pendingDocCchMgr = pendingDocCchMgr
|
|
docPath = docCchMgr.cached()
|
|
currName = f"{docCchMgr.__class__.__name__}: {docCchMgr._identifier}"
|
|
if docPath is not None:
|
|
currName = str(docPath)[6:]
|
|
if currName not in graph:
|
|
graph[currName] = {
|
|
'name': currName,
|
|
'generic_name': str(docCchMgr),
|
|
'type': docCchMgr.__class__.__name__,
|
|
'doc_id': docCchMgr._identifier,
|
|
'monitored': False if docPath is None else docPath.exists(),
|
|
'pub_date': docCchMgr.publication_date(docPath),
|
|
'in_force': docCchMgr.is_in_force(docPath),
|
|
'mention_freq': dict(),
|
|
}
|
|
if docPath in analyzedDocPaths:
|
|
continue
|
|
analyzedDocPaths.append(docPath)
|
|
docFF = DocumentFromFile(str(docPath))
|
|
print(f"Document @ {currName}")
|
|
if docFF is None:
|
|
continue
|
|
doc = docFF.parse()
|
|
newReferences = referenceFinder(doc, docCchMgr.context(docPath))
|
|
if not keep_temporal_context:
|
|
newReferences = list(map(lambda a: a.whithout_temporal_context(), newReferences))
|
|
for newReference in newReferences:
|
|
newDocPath = newReference.cached()
|
|
newName = f"{newReference.__class__.__name__}: {newReference._identifier}"
|
|
if newDocPath is not None:
|
|
newName = str(newDocPath)[6:]
|
|
graph[currName]['mention_freq'][newName] = graph[currName]['mention_freq'].get(newName, 0) + 1
|
|
pendingDocCchMgr = sorted(
|
|
[*pendingDocCchMgr, *newReferences],
|
|
key=lambda dcm: (not dcm.is_cached(), dcm.slowness(), dcm._identifier)
|
|
)
|
|
Path(grapfn).write_text(json.dumps(graph))
|
|
|
|
|
|
def embed_connectivity(graph):
|
|
return graph
|
|
|
|
|
|
def convert_outputs(prefix, temporal_context):
|
|
if not Path(f'{prefix}.json').exists():
|
|
generate_graph(grapfn=f'{prefix}.json', keep_temporal_context=temporal_context)
|
|
graph = json.loads(Path(f'{prefix}.json').read_text())
|
|
if not Path(f'{prefix}_metrics.json').exists():
|
|
Path(f'{prefix}_metrics.json').write_text(json.dumps(embed_connectivity(graph)))
|
|
# to_sqlite
|
|
if Path(f'{prefix}.db').exists():
|
|
Path(f'{prefix}.db').unlink()
|
|
sqldb = sqlite3.connect(f'{prefix}.db')
|
|
cur = sqldb.cursor()
|
|
cur.execute('''CREATE TABLE node (
|
|
name VARCHAR(255),
|
|
type VARCHAR(255),
|
|
doc_id VARCHAR(255),
|
|
monitored bool,
|
|
pub_date VARCHAR(255),
|
|
in_force bool)''')
|
|
cur.execute('''CREATE TABLE edge (
|
|
node_src INTEGER,
|
|
node_dst INTEGER,
|
|
mentions INTEGER,
|
|
FOREIGN KEY(node_src) REFERENCES node(rowid) ON UPDATE CASCADE ON DELETE CASCADE,
|
|
FOREIGN KEY(node_dst) REFERENCES node(rowid) ON UPDATE CASCADE ON DELETE CASCADE)''')
|
|
cur.execute('''CREATE VIEW nodes AS
|
|
SELECT
|
|
rowid as id,
|
|
name as label
|
|
FROM node''')
|
|
cur.execute('''CREATE VIEW edges AS
|
|
SELECT
|
|
rowid as id,
|
|
node_src as source,
|
|
node_dst as target,
|
|
mentions as weight
|
|
FROM edge''')
|
|
node_name_to_id = dict()
|
|
for node in graph.values():
|
|
cur.execute(
|
|
'''INSERT INTO node(name,type,doc_id,monitored,pub_date,in_force) VALUES(?,?,?,?,?,?)''',
|
|
(node['name'], node['type'], node['doc_id'], node['monitored'], node['pub_date'], node['in_force'])
|
|
)
|
|
node_name_to_id[node['name']] = cur.lastrowid
|
|
for node in graph.values():
|
|
node_src_nm = node['name']
|
|
node_src = node_name_to_id[node_src_nm]
|
|
for node_dst_nm, frequency in node['mention_freq'].items():
|
|
node_dst = node_name_to_id[node_dst_nm]
|
|
cur.execute(
|
|
'''INSERT INTO edge(node_src,node_dst,mentions) VALUES(?,?,?)''',
|
|
(node_src, node_dst, frequency)
|
|
)
|
|
cur.close()
|
|
sqldb.commit()
|
|
Path(f'{prefix}.sql').write_text('\n'.join(sqldb.iterdump()))
|
|
sqldb.close()
|
|
# to_csv
|
|
with open(f'{prefix}.csv', 'w') as file:
|
|
file.write('%s,%s,%s\n' % ("source", "target", "weight"))
|
|
for node in graph.values():
|
|
node_src_nm = node['name']
|
|
for node_dst_nm, frequency in node['mention_freq'].items():
|
|
file.write('%s,%s,%d\n' % (node_src_nm, node_dst_nm, frequency))
|
|
# to_graphviz
|
|
gv = graphviz.Digraph()
|
|
for node in graph.values():
|
|
gv.node(
|
|
str(node_name_to_id[node['name']]),
|
|
label='\n'.join(list(map(str, filter(
|
|
lambda a: a is not None,
|
|
[node['type'], node['doc_id'], node['pub_date']]
|
|
))))
|
|
)
|
|
for node in graph.values():
|
|
node_src_nm = node['name']
|
|
node_src = node_name_to_id[node_src_nm]
|
|
for node_dst_nm, frequency in node['mention_freq'].items():
|
|
node_dst = node_name_to_id[node_dst_nm]
|
|
gv.edge(str(node_src), str(node_dst), str(frequency))
|
|
gv.save(f'{prefix}.gv') # takes "forever" to render, "never" finishes
|
|
|
|
|
|
def main():
|
|
convert_outputs('graph', True)
|
|
convert_outputs('graph_noctx', False)
|