def eval_similarity(query_data):
# do actually evaluate similarity ....
query, start_idx, expressions = query_data
csv_reader = csv.reader(expressions,
delimiter='\t',
lineterminator='\n',
quoting=csv.QUOTE_NONE,
escapechar="\\")
end_idx = start_idx + len(expressions) - 1
#create query slt
query_name, query_expression = query
query_tree = SymbolTree.parse_from_slt(query_expression)
query_constraints = Query.create_default_constraints(query_tree)
results = []
for idx, parts in enumerate(csv_reader):
#for idx, expression_info in enumerate(expressions):
#parts = expression_info.strip().split("\t")
expression = parts[0]
doc_id = parts[1]
location = parts[2]
candidate_tree = SymbolTree.parse_from_slt(expression)
try:
data = SIM_FUNCTION(query_tree, candidate_tree, query_constraints)
scores = data[0]
except:
print("Error processing: ")
print(query_expression, flush=True)
print(expression, flush=True)
print("Doc: " + doc_id, flush=True)
print("Loc: " + location, flush=True)
continue
# the index is only returned because some expressions might be absent in case of errors
results.append((scores, start_idx + idx))
print("Processed: " + str(start_idx) + " to " + str(end_idx) + " finished",
flush=True)
return results
def eval_similarity(query_data):
# do actually evaluate similarity ....
query, start_idx, expressions = query_data
end_idx = start_idx + len(expressions) - 1
#create query slt
query_name, query_expression = query
query_tree = SymbolTree.parse_from_slt(query_expression)
query_constraints = Query.create_default_constraints(query_tree)
results = []
for idx, expression_info in enumerate(expressions):
parts = expression_info.strip().split("\t")
expression = parts[0]
doc_id = parts[1]
location = parts[2]
candidate_tree = SymbolTree.parse_from_slt(expression)
try:
scores, matched_q, matched_c, unified_c = similarity_v04(query_tree, candidate_tree, query_constraints)
except:
print("Error processing: ")
print(query_expression, flush=True)
print(expression, flush=True)
print("Doc: " + doc_id, flush=True)
print("Loc: " + location, flush=True)
continue
# the index is only returned because some expressions might be absent in case of errors
results.append((scores, start_idx + idx))
print("Processed: " + str(start_idx) + " to " + str(end_idx) + " finished", flush=True)
return results
def pivot_by_docs(self, how):
# process all query results
"""
how = "core" => use core value ranks directly
"MSS" => use reranking scores
"""
self.by_document = {}
## intID = True # CHANGED TO MATCH ON DOC NAME ALWAYS
if self.tquery:
for doc_id in self.tquery.results.keys():
## try:
## intID = (int(doc_id) == doc_id) # True if doc_id is an integer
## except:
## intID = False # otherwise need to match on filename
(docname, score, positions) = self.tquery.results[doc_id]
# add document if first time seen
# join on docname, not doc_id
try:
doc = self.by_document[docname]
except:
doc = CompQueryResult(doc_id, docname)
self.by_document[docname] = doc
# add score of keyword match to current document
doc.set_tscore(score)
doc.set_tpos(positions)
if self.mqueries:
for qexprnum, query in enumerate(self.mqueries):
# keep scores for all existing formulas over all documents
for result in query.results.values():
# N.B. only one Result structure per matched formula expression
#print("Candidate: " + result.tree.tostring(), flush=True)
if how == "MSS": # compute the MSS score if requested
sim_res = similarity_v06(
query.tree, result.tree,
Query.create_default_constraints(query.tree))
result.new_scores = sim_res[
0] # scores returned as first component of result -- other components are node sets
elif how == "v09":
sim_res = similarity_v09(
query.tree, result.tree,
Query.create_default_constraints(query.tree))
result.new_scores = sim_res[0] # only use scores
elif how == "v10":
sim_res = similarity_v10(
query.tree, result.tree,
Query.create_default_constraints(query.tree))
result.new_scores = sim_res[0] # only use scores
elif how == "v11":
sim_res = similarity_v11(
query.tree, result.tree,
Query.create_default_constraints(query.tree))
result.new_scores = sim_res[0] # only use scores
else:
result.new_scores = [
result.original_score
] # otherwise, just use original score
for doc_id, offset in result.locations:
title = query.documents[doc_id]
#title = title.rpartition('\\')[2] # just last part
title = os.path.basename(title) # just last part (KMD)
## if not intID: # join on title instead of doc_id
joiner = title
## else:
## joiner = doc_id
# add document if first time seen
try:
doc = self.by_document[joiner]
doc.doc_id = doc_id # prefer using math ids (to match positions later)
except:
doc = CompQueryResult(doc_id, title)
self.by_document[joiner] = doc
# add current result to current document
doc.add_mscore(qexprnum, result)
