Python 可视化篇:流式数据监控(转载)

其中:
1.python是2.X版本
2.提供两种实现思路,一是基于matplotlib的animation,一是基于matplotlib的ion
全篇目录为:
一、一点构思
二、matplotlib animation实现思路
  (一)、骨架与实时更新
  (二)、animation的优缺点
三、matplotlib ion实现思路
  (一)、实时更新
  (二)、ion的优缺点
1234567
 
话不多说,先了解大概的效果,如下:
 
 
一、一点构思
       在做此流数据输出可视化前,一直在捣鼓nupic框架,其内部HTM算法主要是一种智能的异常检测算法,是目前AI框架中垂直领域下的一股清流,但由于其实现的例子对应的流数据展示并非我想要的,故此借鉴后自己重新写了一个,主要是达到三个目的,一是展示真实数据的波动,二是展示各波动的异常得分,三是罗列异常的点。
       上述的输出结构并非重点,重点是其实时更新的机制,了解后即可自行定义。另,js对于这种流数据展示应该不难,所以本文主要立足的是算法的呈现角度以及python的实现。
 
二、matplotlib animation实现思路
http://matplotlib.org/api/animation_api.html 链接是matplotlib animation的官方api文档
 
(一)、骨架与实时更新
       animation翻译过来就是动画,其动画展示核心主要有三个:1是动画的骨架先搭好,就是图像的边边框框这些,2是更新的过程,即传入实时数据时图形的变化方法,3是FuncAnimation方法结尾。
       下面以一个小例子做进一步说明:
       1.对于动画的骨架:
 
# initial the figure.
x = []
y = []
fig = plt.figure(figsize=(18, 8), facecolor="white")
ax1 = fig.add_subplot(111)
p1, = ax1.plot(x, y, linestyle="dashed", color="red")123456
以上分别对应初始化空数据,初始化图形大小和背景颜色,插入子图(三个数字分别表示几行几列第几个位置),初始化图形(数据为空)。
 
import numpy as np
x = np.arange(0, 1000, 1)
y = np.random.normal(100, 10, 1000)123
随机生成一些作图数据,下面定义update过程。
       2.对于更新过程:
 
def update(i):
    x.append(xs[i])
    y.append(ys[i])
    ax1.set_xlim(min(x),max(x)+1)
    ax1.set_ylim(min(y),max(y)+1)
    p1.set_data(x,y)
    ax1.figure.canvas.draw()
    return p112345678
上述定义更新函数,参数i为每轮迭代从FuncAnimation方法frames参数传进来的数值,frames参数的指定下文会进一步说,x/y通过相应更新之后,对图形的x/y轴大小做相应的重设,再把数据通过set_data传进图形,注意ax1和p1的区别,最后再把上述的变化通过draw()方法绘制到界面上,返回p1给FuncAnimation方法。
       3.对于FuncAnimation方法:
 
ani = FuncAnimation(fig=fig,func=update,frames=len(xs),interval=1)
plt.show()12
FuncAnimation方法主要是与update函数做交互,将frames参数对应的数据逐条传进update函数,再由update函数返回的图形覆盖FuncAnimation原先的图形,fig参数即为一开始对应的参数,interval为每次更新的时间间隔,还有其他一些参数如blit=True控制图形精细,当界面较多子图时,为True可以使得看起来不会太卡,关键是frames参数,下面是官方给出的注释:
 
可为迭代数,可为函数,也可为空,上面我指定为数组的长度,其迭代则从0开始到最后该数值停止。
       该例子最终呈现的效果如下:
 
了解大概的实现,细节就不在这里多说了。
 
(二)、animation的优缺点
       animation的绘制的结果相比于下文的ion会更加的细腻,主要体现在FuncAnimation方法的一些参数的控制上。但是缺点也是明显,就是必须先有指定的数据或者指定的数据大小,显然这样对于预先无法知道数据的情况没法处理。所以换一种思路,在matplotlib ion打开的模式下,每次往模板插入数据都会进行相应的更新,具体看第二部分。
 
三、matplotlib ion实现思路
(一)、实时更新
       matplotlib ion的实现也主要是三个核心,1是打开ion,2是实时更新机制,3是呈现在界面上。
       1.对于打开ion:
ion全称是 interactive on(交互打开),其意为打开一个图形的交互接口,之后每次绘图都在之前打开的面板上操作,举个例子:
import matplotlib.pyplot as plt
plt.ion()
fig = plt.figure()
ax1 = fig.add_subplot(111)
line, = ax1.plot(t, v, linestyle="-", color="r")12345
打开交互接口,初始化图形。
       2.对于实时更新机制:
import numpy as np
ys = np.random.normal(100, 10, 1000)
def p(a, b):
    t.append(a)
    v.append(b)
    ax1.set_xlim(min(t), max(t) + 1)
    ax1.set_ylim(min(v), max(v) + 1)
    line.set_data(t, v)
    plt.pause(0.001)
    ax1.figure.canvas.draw()
for i in xrange(len(ys)):
    p(i, ys[i])1234567891011121314
随机生成一组数据,定义作图函数p(包含pause表示暂定时延,最好有,防止界面卡死),传入数据实时更新。
       3.对于界面最终呈现
 
plt.ioff()
plt.show()12
ioff是关闭交互模式,就像open打开文件产生的句柄,最好也有个close关掉。
       最终效果如下:

(二)、ion的优缺点
       animation可以在细节上控制比ion更加细腻,这也是ion没有的一点,但是单就无需预先指定数据这一点,ion也无疑是能把流数据做得更加好。
四、最后
       贴一下两种方法在最开始那种图的做法,ion我定义成类,这样每次调用只需穿入参数就可以。
animation版本
# _*_ coding:utf-8 _*_
import os
import csv
import datetime
import matplotlib.pyplot as plt
from matplotlib.animation import FuncAnimation
from matplotlib.dates import DateFormatter
import matplotlib.ticker as ticker
# read the file
filePath = os.path.join(os.getcwd(), "data/anomalyDetect_output.csv")
file = open(filePath, "r")
allData = csv.reader(file)
# skip the first three columns
allData.next()
allData.next()
allData.next()
# cache the data
data = [line for line in allData]
# for i in data: print i
# take out the target value
timestamp = [line[0] for line in data]
value = [line[1:] for line in data]

# format the time style 2016-12-01 00:00:00
def timestampFormat(t):
    result = datetime.datetime.strptime(t, "%Y-%m-%d %H:%M:%S")
    return result

# take out the data
timestamp = map(timestampFormat, timestamp)
value_a = [float(x[0]) for x in value]
predict_a = [float(x[1]) for x in value]
anomalyScore_a = [float(x[2]) for x in value]
# initial the size of the figure
fig = plt.figure(figsize=(18, 8), facecolor="white")
fig.subplots_adjust(left=0.06, right=0.70)
ax1 = fig.add_subplot(2, 1, 1)
ax2 = fig.add_subplot(2, 1, 2)
ax3 = fig.add_axes([0.8, 0.1, 0.2, 0.8], frameon=False)
# initial plot
p1, = ax1.plot_date([], [], fmt="-", color="red", label="actual")
ax1.legend(loc="upper right", frameon=False)
ax1.grid(True)
p2, = ax2.plot_date([], [], fmt="-", color="red", label="anomaly score")
ax2.legend(loc="upper right", frameon=False)
ax2.axhline(0.8, color='black', lw=2)
# add the x/y label
ax2.set_xlabel("date time")
ax2.set_ylabel("anomaly score")
ax1.set_ylabel("value")
# add the table in ax3
col_labels = ["date time", 'actual value', 'predict value', 'anomaly score']
ax3.text(0.05, 0.99, "anomaly value table", size=12)
ax3.set_xticks([])
ax3.set_yticks([])
# axis format
dateFormat = DateFormatter("%m/%d %H:%M")
ax1.xaxis.set_major_formatter(ticker.FuncFormatter(dateFormat))
ax2.xaxis.set_major_formatter(ticker.FuncFormatter(dateFormat))

# define the initial function
def init():
    p1.set_data([], [])
    p2.set_data([], [])
    return p1, p2

# initial data for the update function
x1 = []
x2 = []
x1_2 = []
y1_2 = []
x1_3 = []
y1_3 = []
y1 = []
y2 = []
highlightList = []
turnOn = True
tableValue = [[0, 0, 0, 0]]

# update function
def stream(i):
    # update the main graph(contains actual value and predicted value)
    # add the data
    global turnOn, highlightList, ax3
    x1.append(timestamp[i])
    y1.append(value_a[i])
    # update the axis
    minAxis = max(x1) - datetime.timedelta(days=1)
    ax1.set_xlim(minAxis, max(x1))
    ax1.set_ylim(min(y1), max(y1))
    ax1.figure.canvas.draw()
    p1.set_data(x1, y1)
    # update the anomaly graph(contains anomaly score)
    x2.append(timestamp[i])
    y2.append(anomalyScore_a[i])
    ax2.set_xlim(minAxis, max(x2))
    ax2.set_ylim(min(y2), max(y2))
    # update the scatter
    if anomalyScore_a[i] >= 0.8:
        x1_3.append(timestamp[i])
        y1_3.append(value_a[i])
        ax1.scatter(x1_3, y1_3, s=50, color="black")
    # update the high light
    if anomalyScore_a[i] >= 0.8:
        highlightList.append(i)
        turnOn = True
    else:
        turnOn = False
    if len(highlightList) != 0 and turnOn is False:
        ax2.axvspan(timestamp[min(highlightList)] - datetime.timedelta(minutes=10),
                    timestamp[max(highlightList)] + datetime.timedelta(minutes=10),
                    color='r',
                    edgecolor=None,
                    alpha=0.2)
        highlightList = []
        turnOn = True
    p2.set_data(x2, y2)
    # add the table in ax3
    # update the anomaly tabel
    if anomalyScore_a[i] >= 0.8:
        ax3.remove()
        ax3 = fig.add_axes([0.8, 0.1, 0.2, 0.8], frameon=False)
        ax3.text(0.05, 0.99, "anomaly value table", size=12)
        ax3.set_xticks([])
        ax3.set_yticks([])
        tableValue.append([timestamp[i].strftime("%Y-%m-%d %H:%M:%S"), value_a[i], predict_a[i], anomalyScore_a[i]])
        if len(tableValue) >= 40: tableValue.pop(0)
        ax3.table(cellText=tableValue, colWidths=[0.35] * 4, colLabels=col_labels, loc=1, cellLoc="center")
    return p1, p2

# main animated function
anim = FuncAnimation(fig, stream, init_func=init, frames=len(timestamp), interval=0)
plt.show()
file.close()123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154
ion版本
#! /usr/bin/python
import os
import csv
import datetime
import matplotlib.pyplot as plt
from matplotlib.animation import FuncAnimation
from matplotlib.dates import DateFormatter
import matplotlib.ticker as ticker

class streamDetectionPlot(object):
    """
    Anomaly plot output.
    """
    # initial the figure parameters.
    def __init__(self):
        # Turn matplotlib interactive mode on.
        plt.ion()
        # initial the plot variable.
        self.timestamp = []
        self.actualValue = []
        self.predictValue = []
        self.anomalyScore = []
        self.tableValue = [[0, 0, 0, 0]]
        self.highlightList = []
        self.highlightListTurnOn = True
        self.anomalyScoreRange = [0, 1]
        self.actualValueRange = [0, 1]
        self.predictValueRange = [0, 1]
        self.timestampRange = [0, 1]
        self.anomalyScatterX = []
        self.anomalyScatterY = []
        # initial the figure.
        global fig
        fig = plt.figure(figsize=(18, 8), facecolor="white")
        fig.subplots_adjust(left=0.06, right=0.70)
        self.actualPredictValueGraph = fig.add_subplot(2, 1, 1)
        self.anomalyScoreGraph = fig.add_subplot(2, 1, 2)
        self.anomalyValueTable = fig.add_axes([0.8, 0.1, 0.2, 0.8], frameon=False)
    # define the initial plot method.
    def initPlot(self):
        # initial two lines of the actualPredcitValueGraph.
        self.actualLine, = self.actualPredictValueGraph.plot_date(self.timestamp, self.actualValue, fmt="-",
                                                                  color="red", label="actual value")
        self.predictLine, = self.actualPredictValueGraph.plot_date(self.timestamp, self.predictValue, fmt="-",
                                                                   color="blue", label="predict value")
        self.actualPredictValueGraph.legend(loc="upper right", frameon=False)
        self.actualPredictValueGraph.grid(True)
        # initial two lines of the anomalyScoreGraph.
        self.anomalyScoreLine, = self.anomalyScoreGraph.plot_date(self.timestamp, self.anomalyScore, fmt="-",
                                                                  color="red", label="anomaly score")
        self.anomalyScoreGraph.legend(loc="upper right", frameon=False)
        self.baseline = self.anomalyScoreGraph.axhline(0.8, color='black', lw=2)
        # set the x/y label of the first two graph.
        self.anomalyScoreGraph.set_xlabel("datetime")
        self.anomalyScoreGraph.set_ylabel("anomaly score")
        self.actualPredictValueGraph.set_ylabel("value")
        # configure the anomaly value table.
        self.anomalyValueTableColumnsName = ["timestamp", "actual value", "expect value", "anomaly score"]
        self.anomalyValueTable.text(0.05, 0.99, "Anomaly Value Table", size=12)
        self.anomalyValueTable.set_xticks([])
        self.anomalyValueTable.set_yticks([])
        # axis format.
        self.dateFormat = DateFormatter("%m/%d %H:%M")
        self.actualPredictValueGraph.xaxis.set_major_formatter(ticker.FuncFormatter(self.dateFormat))
        self.anomalyScoreGraph.xaxis.set_major_formatter(ticker.FuncFormatter(self.dateFormat))

    # define the output method.
    def anomalyDetectionPlot(self, timestamp, actualValue, predictValue, anomalyScore):
        # update the plot value of the graph.
        self.timestamp.append(timestamp)
        self.actualValue.append(actualValue)
        self.predictValue.append(predictValue)
        self.anomalyScore.append(anomalyScore)
        # update the x/y range.
        self.timestampRange = [min(self.timestamp), max(self.timestamp)+datetime.timedelta(minutes=10)]
        self.actualValueRange = [min(self.actualValue), max(self.actualValue)+1]
        self.predictValueRange = [min(self.predictValue), max(self.predictValue)+1]
        # update the x/y axis limits
        self.actualPredictValueGraph.set_ylim(
            min(self.actualValueRange[0], self.predictValueRange[0]),
            max(self.actualValueRange[1], self.predictValueRange[1])
        )
        self.actualPredictValueGraph.set_xlim(
            self.timestampRange[1] - datetime.timedelta(days=1),
            self.timestampRange[1]
        )
        self.anomalyScoreGraph.set_xlim(
            self.timestampRange[1]- datetime.timedelta(days=1),
            self.timestampRange[1]
        )
        self.anomalyScoreGraph.set_ylim(
            self.anomalyScoreRange[0],
            self.anomalyScoreRange[1]
        )
        # update the two lines of the actualPredictValueGraph.
        self.actualLine.set_xdata(self.timestamp)
        self.actualLine.set_ydata(self.actualValue)
        self.predictLine.set_xdata(self.timestamp)
        self.predictLine.set_ydata(self.predictValue)
        # update the line of the anomalyScoreGraph.
        self.anomalyScoreLine.set_xdata(self.timestamp)
        self.anomalyScoreLine.set_ydata(self.anomalyScore)
        # update the scatter.
        if anomalyScore >= 0.8:
            self.anomalyScatterX.append(timestamp)
            self.anomalyScatterY.append(actualValue)
            self.actualPredictValueGraph.scatter(
                self.anomalyScatterX,
                self.anomalyScatterY,
                s=50,
                color="black"
            )
        # update the highlight of the anomalyScoreGraph.
        if anomalyScore >= 0.8:
            self.highlightList.append(timestamp)
            self.highlightListTurnOn = True
        else:
            self.highlightListTurnOn = False
        if len(self.highlightList) != 0 and self.highlightListTurnOn is False:
            self.anomalyScoreGraph.axvspan(
                self.highlightList[0] - datetime.timedelta(minutes=10),
                self.highlightList[-1] + datetime.timedelta(minutes=10),
                color="r",
                edgecolor=None,
                alpha=0.2
            )
            self.highlightList = []
            self.highlightListTurnOn = True
        # update the anomaly value table.
        if anomalyScore >= 0.8:
            # remove the table and then replot it
            self.anomalyValueTable.remove()
            self.anomalyValueTable = fig.add_axes([0.8, 0.1, 0.2, 0.8], frameon=False)
            self.anomalyValueTableColumnsName = ["timestamp", "actual value", "expect value", "anomaly score"]
            self.anomalyValueTable.text(0.05, 0.99, "Anomaly Value Table", size=12)
            self.anomalyValueTable.set_xticks([])
            self.anomalyValueTable.set_yticks([])
            self.tableValue.append([
                timestamp.strftime("%Y-%m-%d %H:%M:%S"),
                actualValue,
                predictValue,
                anomalyScore
            ])
            if len(self.tableValue) >= 40: self.tableValue.pop(0)
            self.anomalyValueTable.table(cellText=self.tableValue,
                                         colWidths=[0.35] * 4,
                                         colLabels=self.anomalyValueTableColumnsName,
                                         loc=1,
                                         cellLoc="center"
                                         )
        # plot pause 0.0001 second and then plot the next one.
        plt.pause(0.0001)
        plt.draw()
    def close(self):
        plt.ioff()
        plt.show()
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177
下面是ion版本的调用:
 
graph = stream_detection_plot.streamDetectionPlot()
graph.initPlot()
for i in xrange(len(timestamp)):
    graph.anomalyDetectionPlot(timestamp[i],value_a[i],predict_a[i],anomalyScore_a[i])
graph.close()1234567
具体为实例化类,初始化图形,传入数据作图,关掉。
大概就是这样,有什么不对多指出,谢谢~

原文链接:https://blog.csdn.net/yc_1993/article/details/54933751

猜你喜欢

转载自www.cnblogs.com/xibuhaohao/p/12222673.html