def _parse_select(self, sql):
"""
parsing the sql by the grammar
Select ::= A | AA | AAA | ... |
A ::= agg column table
:return: [Sel(), states]
"""
result = []
select = sql['sql']['select'][1]
result.append(Sel(0))
result.append(N(len(select) - 1))
for sel in select:
result.append(A(sel[0]))
self.colSet.add(sql['col_set'].index(sql['names'][sel[1][1][1]]))
result.append(C(sql['col_set'].index(sql['names'][sel[1][1][1]])))
# now check for the situation with *
if sel[1][1][1] == 0:
result.append(self._parser_column0(sql, select))
else:
result.append(T(sql['col_table'][sel[1][1][1]]))
if not self.copy_selec:
self.copy_selec = [copy.deepcopy(result[-2]), copy.deepcopy(result[-1])]
return result, None
def derive(nodes_type, selected_C):
lst = []
for node_type in nodes_type:
if (node_type == Root1):
node = Root1(3)
elif (node_type == Root):
node = Root(5)
elif (node_type == N):
node = N(0) # 此处存疑，或许也可以是包含selected_A中所有的A
elif (node_type == A):
node = A(0)
elif (node_type == Sel):
node = Sel(0)
elif (node_type == Filter):
id = random.randint(2, 10)
node = Filter(id)
elif (node_type == Order):
id = random.randint(0, 1)
node = Order(id)
elif (node_type == Sup):
id = random.randint(0, 1)
node = Sup(id)
generate(node, selected_C)
lst.append(node)
return lst
def generate(node, selected_C):
if (isinstance(node, A)):
idx = random.randint(0, len(selected_C) - 1)
selected_C[idx].set_parent(node)
node.add_children(selected_C[idx])
return
elif (isinstance(node, Root1)):
child = Root(5)
child.set_parent(node)
node.add_children(child)
generate(child, selected_C)
elif (isinstance(node, Root)):
child = Sel(0)
child.set_parent(node)
node.add_children(child)
generate(child, selected_C)
elif (isinstance(node, Sel)):
child = N(0)
child.set_parent(node)
node.add_children(child)
generate(child, selected_C)
elif (isinstance(node, N) or isinstance(node, Order)
or isinstance(node, Sup) or isinstance(node, Filter)):
child = A(0)
child.set_parent(node)
node.add_children(child)
generate(child, selected_C)
def generate_sketch(node):
if (isinstance(node, N) or isinstance(node, Order)
or isinstance(node, Sup) or isinstance(node, Filter)):
return
elif (isinstance(node, Root1)):
child = Root(5)
child.set_parent(node)
node.add_children(child)
generate_sketch(child)
elif (isinstance(node, Root)):
child = Sel(0)
child.set_parent(node)
node.add_children(child)
generate_sketch(child)
elif (isinstance(node, Sel)):
child = N(0)
child.set_parent(node)
node.add_children(child)
generate_sketch(child)
def _parse_select(self, sql, parent, history_utterance):
"""
parsing the sql by the grammar
Select ::= A | AA | AAA | ... |
A ::= agg column table
:return: [Sel(), states]
"""
result = []
#TODO [[3, [0, [0, 0, False], None]]] [[ 聚合,[0,[0,列id,False,None]]]
select = sql['sql']['select'][1]
result.append(Sel(0, parent))
sub_parent = action_map[Sel]
result.append(N(len(select) - 1, sub_parent))
#TODO select [ [0, [0, [0, 2, False], None]], [0, [0, [0, 6, False], None]] ]
for sel in select:
#TODO sel : [0, [0, [0, 2, False], None]] sel[1][1][1] == 2
#TODO 保存 第一个是聚合函数id,A(none)
result.append(A(sel[0], sub_parent))
#TODO 保存列名
self.colSet.add(sql['col_set'].index(sql['names'][sel[1][1][1]]))
col_id = sql['col_set'].index(sql['names'][sel[1][1][1]])
# now check for the situation with *
#TODO 保存表名 ， 特殊形似，如果是select * ， 那么从sql语句进行推理出属于哪个表。（比较复杂）
if sel[1][1][1] == 0:
table_id = self._parser_column0(sql, select)
if table_id == -1:
col_id = sql['col_set'].index('*')
tc_idx = sql['columns_names_embedder_idxes'].index(
(table_id, col_id))
result.append(TC(tc_idx, sub_parent))
#TODO 类似 SELECT COUNT(*) FROM (SELECT COUNT(*) ... )形式 ，from不是表格
if table_id == -1:
global from_T
full_query = {}
query = {}
full_query['names'] = sql['names']
full_query['col_table'] = sql['col_table']
full_query['col_set'] = sql['col_set']
full_query['table_names'] = sql['table_names']
full_query['keys'] = sql['keys']
full_query['columns_names_embedder'] = sql[
'columns_names_embedder']
full_query['columns_names_embedder_idxes'] = sql[
'columns_names_embedder_idxes']
full_query['nltk_pos'] = sql['nltk_pos']
full_query['utterance_arg_type'] = sql[
'utterance_arg_type']
full_query['utterance_arg'] = sql['utterance_arg']
# print(full_query.keys())
# print(sql.keys())
query['utterance_toks'] = sql['utterance_toks']
query['sql'] = sql['sql']['from']['table_units'][0][1]
# query['query_toks_no_value'] = sql['query_toks_no_value']
query['query'] = sql['query']
query['utterance_arg_type'] = sql['utterance_arg_type']
query['utterance_arg'] = sql['utterance_arg']
from_T += 1
from_result = self.full_parse(full_query, query,
history_utterance)
result.extend(from_result)
else:
table_id = sql['col_table'][sel[1][1][1]]
tc_idx = sql['columns_names_embedder_idxes'].index(
(table_id, col_id))
result.append(TC(tc_idx, sub_parent))
if not self.copy_selec:
self.copy_selec = [
copy.deepcopy(result[-2]),
copy.deepcopy(result[-1])
]
return result, None
for node in new_node_lst:
node.parent = None
node.children = []
return new_node_lst
# print(new_node_lst)
# print(action_p)
if __name__ == '__main__':
# correct_s = "Root1(3) Root(4) Sel(0) N(2) A(0) C(3) T(1) A(0) C(9) T(1) A(0) C(12) T(1) Order(0) A(0) C(12) T(1)".split()
correct = [
Root1(3),
Root(3),
Sel(0),
N(0),
Filter(0),
Filter(0),
Filter(2),
Root(3),
Sel(0),
N(0),
Filter(2),
Filter(2)
]
# predicted_s = 'Root1(3) Root(4) Sel(0) N(2) A(0) C(4)'.split()
predicted = [
Root1(3),
Root(3),
Sel(0),