上文中将数据处理后，要挑出几个重要参数来对比其中的关联性，记录下用pyecharts绘制热力图的流程

上来先处理csv文件，直接复制粘贴

def get_pmu():
    all_pmu_dic = {}
    k8s_wordcount_path = './k8s-100g-micro-1-report.csv'
    with open(k8s_wordcount_path, encoding="utf-8") as f:
        item = np.loadtxt(k8s_wordcount_path, dtype=str, delimiter=',', )
        data = np.loadtxt(k8s_wordcount_path, delimiter=',', skiprows=1)
        for i in range(len(item[0])):
            all_pmu_dic[str(item[0][i])] = {'k8s-wordcount': data[1:, i]}

    yarn_wordcount_path = './yarn-micro-1-report.csv'
    with open(yarn_wordcount_path, encoding='utf-8') as f:
        item = np.loadtxt(yarn_wordcount_path, dtype=str, delimiter=',')
        data = np.loadtxt(yarn_wordcount_path, delimiter=',', skiprows=1)
        for i in range(len(item[0])):
            if not str(item[0][i]) in all_pmu_dic.keys():
                all_pmu_dic[str(item[0][i])] = {'yarn': data[1:, i]}
            else:
                all_pmu_dic[str(item[0][i])].update({'yarn-wordcount': data[1:, i]})

    k8s_terasort_path = './k8s-terasort-micro-1-report.csv'
    with open(k8s_terasort_path, encoding='utf-8') as f:
        item = np.loadtxt(k8s_terasort_path, dtype=str, delimiter=',')
        data = np.loadtxt(k8s_terasort_path, delimiter=',', skiprows=1)
        for i in range(len(item[0])):
            if not str(item[0][i]) in all_pmu_dic.keys():
                all_pmu_dic[str(item[0][i])] = {'k8s-terasort': data[1:, i]}
            else:
                all_pmu_dic[str(item[0][i])].update({'k8s-terasort': data[1:, i]})
    return all_pmu_dic

计算关联性

def correl(matrixA,matrixB):
    matrixA_avg=float(np.mean(matrixA))
    matrixB_avg=float(np.mean(matrixB))
    top_value=float(0)
    down_matrixA=float(0)
    down_matrixB=float(0)
    for i in range(len(matrixA)):
        value=float((matrixA[i]-matrixA_avg)*(matrixB[i]-matrixB_avg))
        top_value=float(top_value+value)
        down_matrixA=float(down_matrixA+np.square(matrixA[i]-matrixA_avg))
        down_matrixB=float(down_matrixB+np.square(matrixB[i]-matrixB_avg))
    down_value=float(np.sqrt(down_matrixB*down_matrixA))
    if not down_value == 0:
        return "%.3f" % (top_value/down_value)
    else:
        return "%.3f" % (float(0))

使用列表记录两个轴的参数

x_axis = ["resource_stalls.sb",
        "msr/pperf/",
        "ld_blocks_partial.address_alias",
        "cache-misses",
        "branch-misses",
        "L1-dcache-load-misses"]
# 我们x轴y轴都一样所以直接赋值
y_axis=x_axis

处理热点图所需的data

data的格式为二维数组,其中二维中要包含x轴坐标、y轴坐标、data，例如[[0,0,1.3],[0,1,2.6],[0,2,1.446]…..]

for x_index in range(len(x_axis)):
    for y_index in range(len(y_axis)):
        index_data=[x_index,y_index]
        matrixA=all_pmu_dic[x_axis[x_index]]['k8s-wordcount']
        matrixB=all_pmu_dic[y_axis[y_index]]['k8s-wordcount']
        correls=float(correl(matrixA,matrixB))
        index_data.append(correls)
        data.append(index_data)

绘制图片

#使用grid在外面包裹一层可以防止属性名太长导致的显示不完全
grid = Grid(init_opts=opts.InitOpts(width="100%",height='950px'))
heatmap = (HeatMap(init_opts=opts.InitOpts())
    .add_xaxis(x_axis) #x轴列表
    .add_yaxis("事件相关性", #
               y_axis, #y轴列表
               data, #数据
               label_opts=opts.LabelOpts(is_show=True, position="inside"), #是否显示每个格子的data
               )
    .set_series_opts()
    .set_global_opts(
        legend_opts=opts.LegendOpts(is_show=True),
        xaxis_opts=opts.AxisOpts(
            type_="category",
            is_scale=True,
            interval=0,
            position="center",
            axislabel_opts=opts.LabelOpts(
                font_size=18,
                interval=0,
                rotate=30,
                border_color="black",
            ),
            splitarea_opts=opts.SplitAreaOpts(
                is_show=True,
                areastyle_opts=opts.AreaStyleOpts(opacity=1),
            ),
        ),
        yaxis_opts=opts.AxisOpts(
            type_="category",
            # is_inverse=True,
            axislabel_opts=opts.LabelOpts(
                font_size=16
            ),
            splitarea_opts=opts.SplitAreaOpts(
                is_show=True, areastyle_opts=opts.AreaStyleOpts(opacity=1)
            ),
        ),
        #相关参数
        title_opts=opts.TitleOpts(title="相关参数相关性",pos_left="15%"),
        #cpu相关参数
        # title_opts=opts.TitleOpts(title="cpu相关性"),
        visualmap_opts=opts.VisualMapOpts(
            min_=float(-1), max_=float(1), is_calculable=True, orient="horizontal", pos_left="center",split_number=5,pos_top=20
        ),
    ))
# heatmap.render(path='cpu.html')
grid.add(heatmap, grid_opts=opts.GridOpts(pos_left="15%",pos_bottom="25%",pos_top="10%"))
grid.render("correl.html") #最后生成名为correl.html的文件

全部代码

import numpy as np
from pyecharts.charts import HeatMap,Grid
from pyecharts import options as opts
from pyecharts.faker import Faker
import random
import seaborn as sns;

def correl(matrixA,matrixB):
    matrixA_avg=float(np.mean(matrixA))
    matrixB_avg=float(np.mean(matrixB))
    top_value=float(0)
    down_matrixA=float(0)
    down_matrixB=float(0)
    for i in range(len(matrixA)):
        value=float((matrixA[i]-matrixA_avg)*(matrixB[i]-matrixB_avg))
        top_value=float(top_value+value)
        down_matrixA=float(down_matrixA+np.square(matrixA[i]-matrixA_avg))
        down_matrixB=float(down_matrixB+np.square(matrixB[i]-matrixB_avg))
    down_value=float(np.sqrt(down_matrixB*down_matrixA))
    if not down_value == 0:
        return "%.3f" % (top_value/down_value)
    else:
        return "%.3f" % (float(0))

def get_pmu():
    all_pmu_dic = {}
    k8s_wordcount_path = './k8s-100g-micro-1-report.csv'
    with open(k8s_wordcount_path, encoding="utf-8") as f:
        item = np.loadtxt(k8s_wordcount_path, dtype=str, delimiter=',', )
        data = np.loadtxt(k8s_wordcount_path, delimiter=',', skiprows=1)
        for i in range(len(item[0])):
            all_pmu_dic[str(item[0][i])] = {'k8s-wordcount': data[1:, i]}

    yarn_wordcount_path = './yarn-micro-1-report.csv'
    with open(yarn_wordcount_path, encoding='utf-8') as f:
        item = np.loadtxt(yarn_wordcount_path, dtype=str, delimiter=',')
        data = np.loadtxt(yarn_wordcount_path, delimiter=',', skiprows=1)
        for i in range(len(item[0])):
            if not str(item[0][i]) in all_pmu_dic.keys():
                all_pmu_dic[str(item[0][i])] = {'yarn': data[1:, i]}
            else:
                all_pmu_dic[str(item[0][i])].update({'yarn-wordcount': data[1:, i]})

    k8s_terasort_path = './k8s-terasort-micro-1-report.csv'
    with open(k8s_terasort_path, encoding='utf-8') as f:
        item = np.loadtxt(k8s_terasort_path, dtype=str, delimiter=',')
        data = np.loadtxt(k8s_terasort_path, delimiter=',', skiprows=1)
        for i in range(len(item[0])):
            if not str(item[0][i]) in all_pmu_dic.keys():
                all_pmu_dic[str(item[0][i])] = {'k8s-terasort': data[1:, i]}
            else:
                all_pmu_dic[str(item[0][i])].update({'k8s-terasort': data[1:, i]})
    return all_pmu_dic

#绘制热力图
#相关参数
x_axis = ["resource_stalls.sb",
        "msr/pperf/",
        "ld_blocks_partial.address_alias",
        "cache-misses",
        "branch-misses",
        "L1-dcache-load-misses"]
#cpu相关参数
# x_axis=["cpu-cycles","cpu-clock","cpu-migrations","cpu_clk_unhalted.ref_tsc","cpu_clk_unhalted.ref_xclk",
# "cpu_clk_unhalted.ref_xclk_any",
# "cpu_clk_unhalted.ring0_trans",
# "cpu_clk_unhalted.thread",
# "cpu_clk_unhalted.thread_any",
# "cpu_clk_unhalted.thread_p",
# "cpu_clk_unhalted.thread_p_any"]
y_axis=x_axis
all_pmu_dic=get_pmu()
data=[]
for x_index in range(len(x_axis)):
    for y_index in range(len(y_axis)):
        index_data=[x_index,y_index]
        matrixA=all_pmu_dic[x_axis[x_index]]['k8s-wordcount']
        matrixB=all_pmu_dic[y_axis[y_index]]['k8s-wordcount']
        correls=float(correl(matrixA,matrixB))
        index_data.append(correls)
        data.append(index_data)


grid = Grid(init_opts=opts.InitOpts(width="100%",height='950px'))
heatmap = (HeatMap(init_opts=opts.InitOpts())
    .add_xaxis(x_axis)
    .add_yaxis("事件相关性",
               y_axis,
               data,
               label_opts=opts.LabelOpts(is_show=True, position="inside"),
               )
    .set_series_opts()
    .set_global_opts(
        legend_opts=opts.LegendOpts(is_show=True),
        xaxis_opts=opts.AxisOpts(
            type_="category",
            is_scale=True,
            interval=0,
            position="center",
            axislabel_opts=opts.LabelOpts(
                font_size=18,
                interval=0,
                rotate=30,
                border_color="black",
            ),
            splitarea_opts=opts.SplitAreaOpts(
                is_show=True,
                areastyle_opts=opts.AreaStyleOpts(opacity=1),
            ),
        ),
        yaxis_opts=opts.AxisOpts(
            type_="category",
            # is_inverse=True,
            axislabel_opts=opts.LabelOpts(
                font_size=16
            ),
            splitarea_opts=opts.SplitAreaOpts(
                is_show=True, areastyle_opts=opts.AreaStyleOpts(opacity=1)
            ),
        ),
        #相关参数
        title_opts=opts.TitleOpts(title="相关参数相关性",pos_left="15%"),
        #cpu相关参数
        # title_opts=opts.TitleOpts(title="cpu相关性"),
        visualmap_opts=opts.VisualMapOpts(
            min_=float(-1), max_=float(1), is_calculable=True, orient="horizontal", pos_left="center",split_number=5,pos_top=20
        ),
    ))
# heatmap.render(path='cpu.html')
grid.add(heatmap, grid_opts=opts.GridOpts(pos_left="15%",pos_bottom="25%",pos_top="10%"))
grid.render("correl.html")