def test_gf(self):
vertices = spark.createDataFrame([('1', 'Carter', 'Derrick', 50),
('2', 'May', 'Derrick', 26),
('3', 'Mills', 'Jeff', 80),
('4', 'Hood', 'Robert', 65),
('5', 'Banks', 'Mike', 93),
('98', 'Berg', 'Tim', 28),
('99', 'Page', 'Allan', 16)],
['id', 'name', 'firstname', 'age'])
edges = spark.createDataFrame([('1', '2', 'friend'),
('2', '1', 'friend'),
('3', '1', 'friend'),
('1', '3', 'friend'),
('2', '3', 'follows'),
('3', '4', 'friend'),
('4', '3', 'friend'),
('5', '3', 'friend'),
('3', '5', 'friend'),
('4', '5', 'follows'),
('98', '99', 'friend'),
('99', '98', 'friend')],
['src', 'dst', 'type'])
g = GraphFrame(vertices, edges)
g.connectedComponents().show()
def main():
# crate spark session
spark = SparkSession.builder.appName("keepindoors graphx connectedComponents()").getOrCreate()
# get a mongo client
cli = mongo.__get__()
# v, ["id","url","title","datetime"]
localVertices=[]
cursor = mongo.getCollection(cli,"keepindoors","docs").find()
for r in cursor:
# del "_id" key which will throws error when createDataFrame
r["id"] = r["docno"]
localVertices.append((r["docno"],r["url"],r["title"],str(r["_id"].generation_time + timedelta(hours=8))))
# e
cursor = mongo.getCollection(cli, "keepindoors", "distances").find()
localEdges = []
for r in cursor:
localEdges.append((r["docno1"],r["docno2"],r["distance"]))
v = spark.createDataFrame(localVertices,["id","url","title","datetime"])
e = spark.createDataFrame(localEdges, ["src", "dst","distance"])
g = GraphFrame(v,e)
# get sparkContext from sparkSession
spark.sparkContext.setCheckpointDir("/tmp/spark/checkpoint")
result = g.connectedComponents()
# order by component,datetime
result = result.orderBy(["component", "datetime"], ascending=[1, 0]).collect()
# create component dict
component_dict = {}
for row in result:
record = row.asDict()
if record["component"] not in component_dict.keys():
component_dict[record["component"]] = []
component_dict[record["component"]].append(record)
# delete mongo collection "component"
mongo.deleteAll(cli,"keepindoors","components")
# save component_dict into mongo
index = 1
for key,item in component_dict.items():
links = []
titles = []
title = "empty title"
update_time = "1970-01-01 00:00:00+00:00"
for doc in item:
titles.append(doc["title"])
links.append(doc["url"])
if doc["datetime"] > update_time:
update_time = doc["datetime"]
title = doc["title"]
mongo.insertDoc({"no":index,"component":key,"title":title,"size":len(item),"links":links,"titles":titles,"update_time":update_time,"docs":item},cli,"keepindoors","components")
index += 1
def get_connected_components(vertices_path, edges_path, checkpoint_dir,
num_reads):
# Read vertices and edges files
df_vertices = build_vertices(vertices_path)
df_edges = build_edges(edges_path, num_reads)
# Build Graph
spark = SparkSession.builder.appName("build_graph").getOrCreate()
vertices = spark.createDataFrame(df_vertices)
edges = spark.createDataFrame(df_edges)
g = GraphFrame(vertices, edges)
# Display Graph
g.vertices.show()
g.edges.show()
# Connected Components
# Get SparkContext using spark.sparkContext
spark.sparkContext.setCheckpointDir(dirName=checkpoint_dir)
result = g.connectedComponents()
dictionary = {}
sorted_result = result.select("id", "component").orderBy('component',
ascending=False)
for row in sorted_result.collect():
if row[1] in dictionary:
dictionary[row[1]].append(row[0])
else:
dictionary[row[1]] = [row[0]]
GL = []
for _, value in dictionary.items():
GL.append(value)
return GL, spark, g
import os
os.environ['PYSPARK_SUBMIT_ARGS'] = '--packages graphframes:graphframes:0.7.0-spark2.4-s_2.11'
import sys
from functools import reduce
from pyspark.sql.functions import col, lit, when
import pyspark
from graphframes.examples import Graphs
from graphframes import GraphFrame
import config
sc = pyspark.SparkContext()
sc.setCheckpointDir('/tmp')
sqlContext = pyspark.SQLContext(sc)
inputFile = sys.argv[1]
# g = Graphs(sqlContext).friends() # Get example graph
df = sqlContext.read.format("csv").option("delimiter", config.delimiter).load(inputFile)
# Rename columns to something decent.
df = df.withColumnRenamed("_c0", "src")\
.withColumnRenamed("_c1", "dst")\
.withColumnRenamed("_c2", "weight")
df.show(5)
aggcodes = df.select("src","dst").rdd.flatMap(lambda x: x).distinct()
vertices = aggcodes.map(lambda x: (x, x)).toDF(["id","name"])
edges = df.select("src", "dst")
graph = GraphFrame(vertices, edges)
result = graph.connectedComponents()
result.select("id", "component").orderBy("component").show()
# COMMAND ----------
stationGraph.bfs(fromExpr="id = 'Townsend at 7th'",
toExpr="id = 'Spear at Folsom'", maxPathLength=2).show(10)
# COMMAND ----------
spark.sparkContext.setCheckpointDir("/tmp/checkpoints")
# COMMAND ----------
minGraph = GraphFrame(stationVertices, tripEdges.sample(False, 0.1))
cc = minGraph.connectedComponents()
# COMMAND ----------
cc.where("component != 0").show()
# COMMAND ----------
scc = minGraph.stronglyConnectedComponents(maxIter=3)
# COMMAND ----------
# of each vertex and returns a graph with each
# vertex assigned a component ID.
# NOTE: With GraphFrames 0.3.0 and later releases,
# the default Connected Components algorithm requires
# setting a Spark checkpoint directory. Users can
# revert to the old algorithm using
# connectedComponents.setAlgorithm("graphx").
#=====================================
# setting a Spark "checkpoint" directory
#=====================================
# What is a Checkpointing? Checkpointing is a process
# of truncating RDD lineage graph and saving it to a
# reliable distributed (HDFS) or local file system.
#
# You call SparkContext.setCheckpointDir(directory: String)
# to set the checkpoint directory - the directory where RDDs
# are checkpointed.
#
spark.sparkContext.setCheckpointDir("/tmp/spark_check_point_dir")
#==========================================
# apply the connectedComponents() algorithm
#==========================================
#
connected_components = graph.connectedComponents()
connected_components.select("id", "component").orderBy("component").show()
# done!
spark.stop()
#filename = '/home/user/leaflet-spark/atom_position_frame_1.npz.npy'
coord_matrix = np.load(filename)
coord_matrix_broadcast = sc.broadcast(coord_matrix)
matrix_size = len(coord_matrix)
dist_Matrix = sc.parallelize(coord_matrix)
dist_Matrix = dist_Matrix.zipWithIndex() #key-value pairs
edge_list = dist_Matrix.flatMap(find_edges)
edge_list = edge_list.filter(lambda x: x[0]!=-1) # filter the -1 values
sqlContext = SQLContext(sc)
Edges = Row('src','dst')
edge = edge_list.map(lambda x: Edges(*x))
e = sqlContext.createDataFrame(edge)
# e.take(10)
v = sqlContext.createDataFrame(sc.parallelize(xrange(matrix_size)).map(lambda i:Row(id=i+1)))
# v.show()
# create the graph
g = GraphFrame(v, e)
#g.vertices.show()
#g.edges.show()
total_time = time() - start_time
cc = g.connectedComponents()
print cc.select("id", "component").orderBy("component").show()
print 'Total time to create the Graphframe: %i sec' % (total_time)
print 'Time to calculate the connected components: %i sec ' % (time() - total_time-start_time)
chain4 = g.find("(a)-[ab]->(b); (b)-[bc]->(c); (c)-[cd]->(d)")
# chain4.show()
# g.find("(c)-[m]->()").show()
# Query on sequence, with state (cnt)
# (a) Define method for updating state given the next element of the motif.
sumFriends = \
lambda cnt, relationship: F.when(relationship == "friend", cnt + 1).otherwise(cnt)
# (b) Use sequence operation to apply method to sequence of elements in motif.
# In this case, the elements are the 3 edges.
condition = \
reduce(lambda cnt, e: sumFriends(cnt, F.col(e).relationship), ["ab", "bc", "cd"], F.lit(0))
# (c) Apply filter to DataFrame.
chainWith2Friends2 = chain4.where(condition >= 2)
# chainWith2Friends2.show()
result = g.connectedComponents()
# 结果含义的解释
# result.show()
# 强连通图和连通图的区别
'''
Connected is usually associated with undirected graphs (two way edges): there is a path between every two nodes.
Strongly connected is usually associated with directed graphs (one way edges): there is a route between every two nodes.
Complete graphs are undirected graphs where there is an edge between every pair of nodes.
'''
result = g.stronglyConnectedComponents(maxIter=10)
# result.orderBy("component").show()
## 社区发现，本质是个聚类算法
result = g.labelPropagation(maxIter=5)
# result.show()