大家好，我是小伍哥，最近在做新的课程《风控策略自动化挖掘》，过程中要用到流程图，发现一个很好用的库pygraphviz，非常好用，我也只是探索了部分功能，这里分享给大家。课件和代码基本完成，视频还在录制过程中，目前小范围开始卖了，看文章就能跑代码，有需要的联系我。
一、导入模块
import pygraphviz as pgv二、创建图形
G = pgv.AGraph(directed=True, strict=False, nodesep=0,ranksep=1.2, rankdir="TB",splines="none", concentrate=True, bgcolor="write",compound=True, normalize=False, encoding='UTF-8')
参数解释
directed -> False | True：有向图
strict -> True | False：简单图
nodesep：同级节点最小间距
ranksep：不同级节点最小间距
rankdir：绘图方向，可选 TB (从上到下), LR (从左到右), BT (从下到上), RL (从右到左)
splines：线条类型，可选 ortho (直角), polyline (折线), spline (曲线), line (线条), none (无)
concentrate -> True | False：合并线条 (双向箭头)
bgcolor：背景颜色
compound -> True | False：多子图时，不允许子图相互覆盖
normalize -> False | True：以第一个节点作为顶节点
三、添加节点
G.add_node(name, label=None, fontname="Times-Roman", fontsize=14,shape="ellipse", style="rounded", color="black", fontcolor="black",pos="x,y(!)", fixedsize=False, width=1, height=1)G.add_nodes_from(names, **attr) # 批量添加点，参数同上
参数解释
name -> str：节点名。label 为节点标签，未指定时显示 name
fontname：字体名称，常用：Microsoft YaHei, SimHei, KaiTi, SimSun, FangSong, Times-Roman, Helvetica, Courier。可以使用 "times bold italic" 表示字体类型、粗细、倾斜
fixedsize -> Flase | True | "shape"`：固定大小，默认随文本长度变化。设置为 True 时，width 和 height 参数共同控制点大小。设置为 "shape" 时，将取标签文本和设置值的较大者
style：节点线样式，使用 `color` 设置线条颜色 (style="filled" 时，设置填充颜色)
shape：节点形状
pos -> str：点的初始位置。使用 "x,y!" 时，可以强制固定点的位置。
四、 添加边
G.add_edge(origin, target, color="black", style="solid",penwidth=1,label="", fontname="Times-Roman", fontsize=14,fontcolor="black",arrowsize=1, arrowhead="normal", arrowtail="normal",dir="forward")G.add_nodes_from([[origin_1, target_1],[origin_2, target_2],...], **attr) # 批量添加线，参数同
参数解释
label -> str：边标签，未指定时不显示
penwidth：线条粗细
arrowsize：箭头大小
arrowhead：箭头类型，可选 normal, vee
dir：箭头方向，可选 both, forward, back, none。只有在无向图中才起作用！
五、导出图形
G.layout()
G.draw(file_name, prog="dot")
prog：布局算法，可选 neato, dot (推荐), twopi, circo, fdp
dot 默认布局方式，主要用于有向图
neato 基于spring-model(又称force-based)算法
twopi 径向布局
circo 圆环布局
fdp 用于无向图
六、案例示例
import graphvizu = graphviz.Digraph('unix', filename='unix.gv',node_attr={'color': 'lightblue2', 'style': 'filled'})u.attr(size='6,6')u.edge('5th Edition', '6th Edition')u.edge('5th Edition', 'PWB 1.0')u.edge('6th Edition', 'LSX')u.edge('6th Edition', '1 BSD')u.edge('6th Edition', 'Mini Unix')u.edge('6th Edition', 'Wollongong')u.edge('6th Edition', 'Interdata')u.edge('Interdata', 'Unix/TS 3.0')u.edge('Interdata', 'PWB 2.0')u.edge('Interdata', '7th Edition')u.edge('7th Edition', '8th Edition')u.edge('7th Edition', '32V')u.edge('7th Edition', 'V7M')u.edge('7th Edition', 'Ultrix-11')u.edge('7th Edition', 'Xenix')u.edge('7th Edition', 'UniPlus+')u.edge('V7M', 'Ultrix-11')u.edge('8th Edition', '9th Edition')u.edge('1 BSD', '2 BSD')u.edge('2 BSD', '2.8 BSD')u.edge('2.8 BSD', 'Ultrix-11')u.edge('2.8 BSD', '2.9 BSD')u.edge('32V', '3 BSD')u.edge('3 BSD', '4 BSD')u.edge('4 BSD', '4.1 BSD')u.edge('4.1 BSD', '4.2 BSD')u.edge('4.1 BSD', '2.8 BSD')u.edge('4.1 BSD', '8th Edition')u.edge('4.2 BSD', '4.3 BSD')u.edge('4.2 BSD', 'Ultrix-32')u.edge('PWB 1.0', 'PWB 1.2')u.edge('PWB 1.0', 'USG 1.0')u.edge('PWB 1.2', 'PWB 2.0')u.edge('USG 1.0', 'CB Unix 1')u.edge('USG 1.0', 'USG 2.0')u.edge('CB Unix 1', 'CB Unix 2')u.edge('CB Unix 2', 'CB Unix 3')u.edge('CB Unix 3', 'Unix/TS++')u.edge('CB Unix 3', 'PDP-11 Sys V')u.edge('USG 2.0', 'USG 3.0')u.edge('USG 3.0', 'Unix/TS 3.0')u.edge('PWB 2.0', 'Unix/TS 3.0')u.edge('Unix/TS 1.0', 'Unix/TS 3.0')u.edge('Unix/TS 3.0', 'TS 4.0')u.edge('Unix/TS++', 'TS 4.0')u.edge('CB Unix 3', 'TS 4.0')u.edge('TS 4.0', 'System V.0')u.edge('System V.0', 'System V.2')u.edge('System V.2', 'System V.3')u.view()
import graphvizf = graphviz.Digraph('finite_state_machine', filename='fsm.gv')f.attr(rankdir='LR', size='8,5')f.attr('node', shape='doublecircle')f.node('LR_0')f.node('LR_3')f.node('LR_4')f.node('LR_8')f.attr('node', shape='circle')f.edge('LR_0', 'LR_2', label='SS(B)')f.edge('LR_0', 'LR_1', label='SS(S)')f.edge('LR_1', 'LR_3', label='S($end)')f.edge('LR_2', 'LR_6', label='SS(b)')f.edge('LR_2', 'LR_5', label='SS(a)')f.edge('LR_2', 'LR_4', label='S(A)')f.edge('LR_5', 'LR_7', label='S(b)')f.edge('LR_5', 'LR_5', label='S(a)')f.edge('LR_6', 'LR_6', label='S(b)')f.edge('LR_6', 'LR_5', label='S(a)')f.edge('LR_7', 'LR_8', label='S(b)')f.edge('LR_7', 'LR_5', label='S(a)')f.edge('LR_8', 'LR_6', label='S(b)')f.edge('LR_8', 'LR_5', label='S(a)')f.view()
多特征策略挖掘流程
Rule ='多特征策略'FeaA = '连续特征'FeaB = '类别特征'FeaC = '文本特征'FeaD = '图片特征'FeaE = '音频特征'method1 = '等宽分箱'method2 = '等频分箱'method3 = '决策树分箱'method4 = '卡方分箱'method5 = '聚类分箱'method6 = '分词等'method7 = '目标检测'method8 = '音转文等'stat = '统计计算'rest = '风控策略'Combine = '特征组合'nodes = [FeaA, FeaB,FeaC,FeaD,FeaE,stat, rest, method1, method2, method3, method4,method5,method6,method7,method8,combine]import pygraphviz as pgvG = pgv.AGraph(directed=True,strict=False,ranksep=0.46,nodesep=0.42,splines="ortho",rankdir='LR',concentrate=True)G.add_node(Rule,color="#DDA0DD",style="filled",fontname="times bold italic" ,shape='egg')G.add_nodes_from(nodes, color="#87CEEB",style="filled",fontname="times bold italic")G.add_edges_from([[Rule, FeaA],[],[],[],[],[],[],[],[],[],[],[],[],[],[],[],[method2, Combine], [method3, Combine], [method4, Combine],[],[method6,Combine],[method7,Combine],[method8,Combine],[],[stat,rest]],color="#7F01FF", arrowsize=0.8)G.graph_attr.update(dpi='800')G.layout()G.draw("多特征策略.png", prog="dot")from IPython.display import Imagedisplay(Image(filename="多特征策略.png"))
风控策略总体流程
分箱方法
因果关系图
import pygraphviz as pgvG = pgv.AGraph(directed=True, strict=False, ranksep=0.2, splines="spline", concentrate=True)# 设置节点标签nodeA = "Police\nIntelligence"nodeB = "Police Station"nodeC = "Criminal Action"nodeD = "Incidents"nodeE = "Police Dockets"nodeF = "Control Room\nAwareness"nodeG = "Patroller Information"nodeH = "Patroller Awareness"# 添加节点G.add_nodes_from([nodeA, nodeB, nodeC, nodeD, nodeE, nodeF, nodeG, nodeH],color="#ffffff", fontname="times bold italic")# 添加边G.add_edges_from([[nodeA, nodeB], [nodeA, nodeF], [nodeB, nodeC], [nodeC, nodeD],[nodeC, nodeG], [nodeD, nodeE], [nodeD, nodeG], [nodeE, nodeA],[nodeF, nodeA], [nodeF, nodeG], [nodeF, nodeH], [nodeG, nodeF],[nodeH, nodeG]], color="#7F01FF", arrowsize=0.8)# 导出图形G.layout()G.draw("因果关系图.png", prog="dot")
因子相关性图
import pygraphviz as pgvG = pgv.AGraph(directed=True, rankdir="TB")# 设置节点标签Root = "道路交通流畅"negative_1 = "平均延误时间"negative_2 = "负荷度"negative_3 = "小区位置"negative_4 = "相对延误率"negative_5 = "房屋密度"negative_6 = "人口密度"negative_7 = "总延误率"negative_8 = "排队率"negative_9 = "行驶时间"positive_1 = "通行能力"positive_2 = "公路层级"positive_3 = "路网结构"positive_4 = "行驶速度"positive_5 = "路网长度"positive_6 = "小区面积"positive_7 = "内部道路密度"positive_8 = "路网密度"# 添加节点G.add_node(Root, style="filled", shape="box3d", color="#feb64d")for negative in [eval(_) for _ in dir() if _.startswith("negative")]:G.add_node(negative, style="filled", shape="ellipse", color="#CFDBF6")for positive in [eval(_) for _ in dir() if _.startswith("positive")]:G.add_node(positive, style="filled", shape="ellipse", color="#B4E7B7")# 添加边G.add_edges_from([[Root, negative_1], [Root, negative_6], [Root, negative_8], [Root, negative_9],[negative_1, negative_2], [negative_1, negative_7], [negative_2, negative_3],[negative_2, negative_7], [negative_3, negative_4], [negative_8, negative_9],[positive_2, negative_5], [positive_3, negative_4], [positive_4, negative_5]],color="#B4DBFF", style="dashed", penwidth=1.5)G.add_edges_from([[Root, positive_1], [Root, positive_8], [negative_5, negative_4],[negative_6, positive_4], [negative_5, positive_4], [negative_9, positive_5],[positive_1, positive_2], [positive_2, positive_3], [positive_6, positive_5],[positive_7, positive_6], [positive_8, positive_7]],color="#B4E7B7", style="dashed", penwidth=1.5)# 导出图形G.layout()G.draw("因子相关性图.png", prog="dot")
更复杂的案例
往期精彩：
SynchroTrap-基于松散行为相似度的欺诈账户检测算法
