Elasticsearch(GEO)空间检索查询
Elasticsearch(GEO)空间检索查询python版本
1、Elasticsearch
ES的强大就不用多说了，当你安装上插件，搭建好集群，你就拥有了一个搜索系统。
当然，ES的集群优化和查询优化就是另外一个议题了。这里mark一个最近使用的es空间检索的功能。
2、ES GEO空间检索
空间检索顾名思义提供了通过空间距离和位置关系进行检索的能力。有很多空间索引算法和类库可供选择。
ES内置了这种索引方式。下面详细介绍。
step1：创建索引
def create_index(): mapping = { "mappings": { "poi": { "_routing": { "required": "true", "path": "city_id" }, "properties": { "id": { "type": "integer" }, "geofence_type": { "type": "integer" }, "city_id": { "type": "integer" }, "city_name": { "type": "string", "index": "not_analyzed" }, "activity_id": { "type": "integer" }, "post_date": { "type": "date" }, "rank": { "type": "float" }, # 不管是point还是任意shape, 都用geo_shape,通过type来设置 # type在数据里 "location_point": { "type": "geo_shape" }, "location_shape": { "type": "geo_shape" }, # 在计算点间距离的时候, 需要geo_point类型变量 "point": { "type": "geo_point" } } } } } # 创建索引的时候可以不 mapping es.create_index(index='mapapp', body=mapping) # set_mapping = es_dsl.set_mapping('mapapp', 'poi', body=mapping)
这里我们创建了一个名叫mapapp的索引，映射的设置如mapping所示。
2、批量插入数据bulk
def bulk():
# actions 是一个可迭代对象就行, 不一定是list
workbooks = xlrd.open_workbook('./geo_data.xlsx')
table = workbooks.sheets()[1]
colname = list()
actions = list()
for i in range(table.nrows):
if i == 0:
colname = table.row_values(i)
continue
geo_shape_point = json.loads(table.row_values(i)[7])
geo_shape_shape = json.loads(table.row_values(i)[8])
geo_point = json.loads(table.row_values(i)[9])
raw_data = table.row_values(i)[:7]
raw_data.extend([geo_shape_point, geo_shape_shape, geo_point])
source = dict(zip(colname, raw_data))
geo = GEODocument(**source)
action = {
"_index": "mapapp",
"_type": "poi",
"_id": table.row_values(i)[0],
"_routing": geo.city_id,
#"_source": source,
"_source": geo.to_json(),
}
actions.append(action)
es.bulk(index='mapapp', actions=actions, es=es_handler, max=25)
刷入测试数据，geo_data数据形如：
id geofence_type city_id city_name activity_id post_date rank location_point location_shape point 1 1 1 北京 100301 2016/10/20 100.30 {"type":"point","coordinates":[55.75,37.616667]} {"type":"polygon","coordinates":[[[22,22],[4.87463,52.37254],[4.87875,52.36369],[22,22]]]} {"lat":55.75,"lon":37.616667} 2 1 1 北京 100302 2016/10/21 12.00 {"type":"point","coordinates":[55.75,37.616668]} {"type":"polygon","coordinates":[[[0,0],[4.87463,52.37254],[4.87875,52.36369],[0,0]]]} {"lat":48.8567,"lon":2.3508} 3 1 1 北京 100303 2016/10/22 3432.23 {"type":"point","coordinates":[55.75,37.616669]} {"type":"polygon","coordinates":[[[4.8833,52.38617],[4.87463,52.37254],[4.87875,52.36369],[4.8833,52.38617]]]} {"lat":32.75,"lon":37.616668} 4 1 1 北京 100304 2016/10/23 246.80 {"type":"point","coordinates":[52.4796, 2.3508]} {"type":"polygon","coordinates":[[[4.8833,52.38617],[4.87463,52.37254],[4.87875,52.36369],[4.8833,52.38617]]]} {"lat":11.56,"lon":37.616669}
3、GEO查询：两点间距离
# 点与点之间的距离 # 按照距离升序排列,如果size取1个,就是最近的 def sort_by_distance(): body = { "from": 0, "size": 1, "query": { "bool": { "must": [{ "term": { "geofence_type": 1 } }, { "term": { "city_id": 1 } }] } }, "sort": [{ "_geo_distance": { "point": { "lat": 8.75, "lon": 37.616 }, "unit": "km", "order": "asc" } }] } for i in es.search(index='mapapp', doc_type='poi', body=body)['hits']['hits']: print type(i), i
4、GEO查询：边界框过滤
tips：大家都知道，ES的过滤是会生成缓存的，所以在优化查询的时候，常常需要将频繁用到的查询提取出来作为过滤呈现，但不幸的是，对于GEO过滤不会生成缓存，所以没有必要考虑，这里为了做出区分，使用post_filter，查询后再过滤，下面的都类似。
# 边界框过滤:用框去圈选点和形状 # 这里实现了矩形框选中 # post_filter后置filter, 对查询结果再过滤; aggs常用后置filter def bounding_filter(): body = { "from": 0, "size": 1, "query": { "bool": { "must": [{ "term": { "geofence_type": 1 } }, { "term": { "city_id": 1 } }] } }, "post_filter": { "geo_shape": { "location_point": { "shape": { "type": "envelope", "coordinates": [[52.4796, 2.3508], [48.8567, -1.903]] }, "relation": "within" } } } } for i in es.search(index='mapapp', doc_type='poi', body=body)['hits']['hits']: print type(i), i
5、GEO查询：圆形圈选
# 边界框过滤: 圆形圈选 # post_filter后置filter, 对查询结果再过滤; aggs常用后置filter def circle_filter(): body = { "from": 0, "size": 1, "query": { "bool": { "must": [{ "term": { "geofence_type": 1 } }, { "term": { "city_id": 1 } }] } }, "post_filter": { "geo_shape": { "location_point": { "shape": { "type": "circle", "radius": "10000km", "coordinates": [22, 45] }, "relation": "within" } } } } for i in es.search(index='mapapp', doc_type='poi', body=body)['hits']['hits']: print type(i), i
6、GEO查询：反选
# 边界框反选:点落在框中,框被查询出来 # post_filter后置filter, 对查询结果再过滤; aggs常用后置filter # 包含正则匹配regexp def intersects(): body = { "from": 0, "size": 1, "query": { "bool": { "must": [{ "term": { "geofence_type": 1 } }, { "regexp": { "city_name": u".*北京.*" } }, { "term": { "city_id": 1 } }] } }, "post_filter": { "geo_shape": { "location_shape": { "shape": { "type": "point", "coordinates": [22,22] }, "relation": "intersects" } } } } for i in es.search(index='mapapp', doc_type='poi', body=body)['hits']['hits']: print type(i), i
7、最后粘两个空间聚合的例子，作为参考
# 空间聚合 # 按照与中心点距离聚合 def aggs_geo_distance(): body = { "aggs": { "aggs_geopoint": { "geo_distance": { "field": "point", "origin": { "lat": 51.5072222, "lon": -0.1275 }, "unit": "km", "ranges": [ { "to": 1000 }, { "from": 1000, "to": 3000 }, { "from": 3000 } ] } } } } for i in es.search(index='mapapp', doc_type='poi', body=body)['aggregations']['aggs_geopoint']['buckets']: print type(i), i # 空间聚合 # geo_hash算法, 网格聚合grid # 两次聚合 def aggs_geohash_grid(): body = { "aggs": { "new_york": { "geohash_grid": { "field": "point", "precision": 5 } }, "map_zoom": { "geo_bounds": { "field": "point" } } } } for i in es.search(index='mapapp', doc_type='poi', body=body)['aggregations']['new_york']['buckets']: print type(i), i
