全网最全的平行坐标图(parallel coordinates plot)的绘制攻略

早上起来拥抱太阳，写小论文，看到人家的图怎么那么好看！！？？ 这不得赶紧抄下来，我也发一个顶刊？于是开始思考如何解决绘制这个问题，目前现有的大部分解决方案都是直接调库，查了一下现有的所有解决方案，

1.直接python调库

这篇文章直接调库，但是没有分类，而且也感觉不太好看

https://www.zhihu.com/question/571887313

这一个是调pandas库的，虽然有分类但是他的轴不能归一化调整。

https://www.cnblogs.com/caiyishuai/p/12322671.html

2.matlab

matlab作为强大的科研工具是众所周知的，他也提供了一个库来专门绘制平行坐标图。但是由于电脑内存不够，我就直接用线上matlab。经过学习之后，绘制出来这样的图，emmm老实说有点丑，但是指不定有人需要这个呢？

tbl = readtable('tt.csv');

head(tbl);

tbl.Name = categorical(tbl.Name);

p = parallelplot(tbl);

p.GroupVariable = 'Name';

p.CoordinateVariables = [3 2 4];

p.LineWidth = 1.5;

p.FontSize = 25;

p.Color = {'#B4FF00','#00E68C','#1432FA','#B4C8FA','#FF3232'}

%p.Color = {'#780001','#C11221','#FEF0D5','#002F49','#669BBB'}

其中csv的格式大概是这样的

cost time energy Name 0 2 20 GMPSO 0 8 90 KAMSA 6 5 30 COSA

太丑了还是下一个

3.用Origin绘制

Origin不愧是专业的绘图工具，绘制出来的图确实还不错。这里也是因为电脑内存原因所以我就没试着用Origin了。贴个参考文献：

https://cloud.tencent.com/developer/article/1623006?areaSource=102001.5&traceId=e-JyHo2xQfKU1fPJYmtbA

4.用高手做的轮子

上面尽管提供了三种方案，但是感觉也不能绘制出我想要的图形。于是乎我就上github寻找，肯定有大神。这个大神是我目前找到最满意的解决方案了。

https://github.com/jraine/parallel-coordinates-plot-dataframe

这个仓库提供了一个不错的解决方案，他能绘制出好看的平行坐标图，而且也不用引太多的库。

照着这个代码魔改了一下，我就绘制出来这样的图形了 这不比上面的图要好看？我就是天才啊哈哈哈，现在先问问老师我最后一格要不要换成图例，不用的话论文就贴这个图了。下面是魔改后的代码：

import matplotlib

from matplotlib import ticker

import matplotlib.pyplot as plt

import numpy as np

import pandas as pd

from scipy.interpolate import make_interp_spline

def read_and_add_method(file_path, method_name):

df = pd.read_csv(file_path)

if method_name == 'ChaoticOSA':

method_name = 'COSA'

if method_name == 'WDNMN':

method_name = 'WDNS'

df['method'] = method_name # 添加新列记录算法名

return df

def parallel_plot(df,cols,rank_attr,cmap='Spectral',spread=False,curved=0.1,curvedextend=0.05):

'''Produce a parallel coordinates plot from pandas dataframe with line colour with respect to a column.

Required Arguments:

df: dataframe

cols: columns to use for axes

rank_attr: attribute to use for ranking

Options:

cmap: Colour palette to use for ranking of lines

spread: Spread to use to separate lines at categorical values

curved: Spline interpolation along lines

curvedextend: Fraction extension in y axis, adjust to contain curvature

Returns:

x coordinates for axes, y coordinates of all lines'''

colmap = matplotlib.cm.get_cmap(cmap)

cols = cols + [rank_attr]

fig, axes = plt.subplots(1, len(cols)-1, sharey=False, figsize=(3*len(cols)+3,5))#绘制三个子图

valmat = np.ndarray(shape=(len(cols),len(df)))#定义需要绘制曲线的数组有df行，cols列

x = np.arange(0,len(cols),1)#貌似没什么用,有3列那么x=[0,1,2]

ax_info = {}

for i,col in enumerate(cols):#归一化数据

vals = df[col]

if (vals.dtype == float) & (len(np.unique(vals)) > 20):

minval = np.min(vals)

maxval = np.max(vals)

rangeval = maxval - minval#区间长度

vals = np.true_divide(vals - minval, maxval-minval)#归一化处理vals-minval/maxval-minval除法运算

nticks = 5

tick_labels = [round(minval + i*(rangeval/nticks),4) for i in range(nticks+1)]

ticks = [0 + i*(1.0/nticks) for i in range(nticks+1)]

valmat[i] = vals

ax_info[col] = [tick_labels,ticks]

else:

vals = vals.astype('category')#假如是目录型

cats = vals.cat.categories

c_vals = vals.cat.codes

minval = 0

maxval = len(cats)-1

if maxval == 0:

c_vals = 0.5

else:

c_vals = np.true_divide(c_vals - minval, maxval-minval)

tick_labels = cats

ticks = np.unique(c_vals)

ax_info[col] = [tick_labels,ticks]

if spread is not None:

offset = np.arange(-1,1,2./(len(c_vals)))*2e-2

np.random.shuffle(offset)

c_vals = c_vals + offset

valmat[i] = c_vals

extendfrac = curvedextend if curved else 0.05

for i,ax in enumerate(axes):

for idx in range(valmat.shape[-1]):

if curved:

x_new = np.linspace(0, len(x), len(x)*20)

a_BSpline = make_interp_spline(x, valmat[:,idx],k=3,bc_type='clamped')

y_new = a_BSpline(x_new)

ax.plot(x_new,y_new,color=colmap(valmat[-1,idx]),alpha=0.5)

else:

ax.plot(x,valmat[:,idx],color=colmap(valmat[-1,idx]),alpha=0.5)

ax.set_ylim(0-extendfrac,1+extendfrac)

ax.set_xlim(i,i+1)

for dim, (ax,col) in enumerate(zip(axes,cols)):

ax.xaxis.set_major_locator(ticker.FixedLocator([dim]))

ax.yaxis.set_major_locator(ticker.FixedLocator(ax_info[col][1]))

ax_info[col][0] = [int(label) for label in ax_info[col][0]]#y标签下取整

ax.set_yticklabels(ax_info[col][0])

ax.set_xticklabels([cols[dim]])

plt.subplots_adjust(wspace=0)

norm = matplotlib.colors.Normalize(0,1)#*axes[-1].get_ylim())

sm = plt.cm.ScalarMappable(cmap=cmap, norm=norm)

cbar = plt.colorbar(sm,pad=0,ticks=ax_info[rank_attr][1],extend='both',extendrect=True,extendfrac=extendfrac)

#if curved:

#cbar.ax.set_ylim(0-curvedextend,1+curvedextend)

cbar.ax.set_yticklabels(ax_info[rank_attr][0])

cbar.ax.set_xlabel(rank_attr)

plt.show()

return x,valmat

method_names = ['GALCS','GMPSO',"ChaoticOSA","KAMSA","WDNMN"]

data_size = 30

data_index = 50

data_set_name = 'CyberShake'

# 存储所有DataFrame的列表

all_dataframes = []

# 遍历目录下所有的CSV文件

for method_name in method_names:

fileName = 'D://demo//dataset/5.0-5.0/'+str(data_size)+'/'+str(data_size)+data_set_name+str(data_index)+method_name+'_0.csv'

df = read_and_add_method(fileName, method_name)

# 将DataFrame添加到列表中

all_dataframes.append(df)

# 将所有的DataFrame拼接在一起

final_dataframe = pd.concat(all_dataframes, ignore_index=True)

print(final_dataframe)

parallel_plot(final_dataframe,['time','cost','energy'],'method')

# 定义函数，读取CSV文件并添加一个新列'method'

解决了一个问题咯，拜拜咯