(进阶)003 - AWS DeepRacer分析log

下载 Evaluation log

解压log，找到 0-iteration.csv

log 可视化分析

import os
import numpy as np
import pandas as pd
from math import exp
from matplotlib.collections import LineCollection
import matplotlib.pyplot as plt
import matplotlib.animation as animation


def getCSVdata(file):
    with open(file,'r') as fo:
        return fo.read()

def getPathXY():
    path = pd.read_csv('Invent 2018.csv',header=None)
    pathList = path.values.tolist()
    X,Y = [item[0]*100 for item in pathList],[item[1]*100 for item in pathList]
    return X,Y
def getPDInfo(excelname,sheetname):
    dataAll = pd.read_excel(excelname,sheet_name=sheetname)
    timeSet = set(dataAll.episode.tolist())
    return dataAll,timeSet
def getTimesXYSBySheetName(dataAll,timeIndex):
    data = dataAll[dataAll.episode == timeIndex]
    x,y,tstamp,colorListSelectSpeed,colorListSelectReward,speedL,rewardList = [x*100 for x in data.X.values.tolist()],[y*100 for y in data.Y.values.tolist()],[t*0.3 for t in data.tstamp.values.tolist()],[t for t in data.color.values.tolist()][1:],[t for t in data.Rcolor.values.tolist()][1:],[t for t in data.throttle.values.tolist()][1:],[t for t in data.reward.values.tolist()][1:]
    tstampDelta = [tstamp[i+1]-tstamp[i] for i in range(len(tstamp)-1)]
    return x,y,tstampDelta,colorListSelectSpeed,colorListSelectReward,speedL,rewardList
def DrawTrack(fig,ex,ey,tstamp,colorListSelectSpeed,colorListSelectReward,speedL,rewardList):  #39  0.387   63.43
    ims = []
    for i in range(len(ex)-1):
        plt.title('speed:'+str(round(speedL[i],2)) + '\nreward:' + str(round(rewardList[i],3)))
        plt.xlabel('Env: re:Invent 2018\nunite:cm')
        im = plt.plot(ex[0:i+1],ey[0:i+1],color=colorListSelectSpeed[i],linewidth=4)
        ims.append(im)
        # plt.plot(xs, ys,color=color,linewidth=12.365) # 车宽算出来12.365cm
        plt.pause(tstamp[i])
    # ani = animation.ArtistAnimation(fig, ims, repeat_delay=1000)
    # ani.save("1.gif", writer='pillow')
    return ims
def speed2Color(speed,speedList,colorList):
    for index,p in enumerate(speedList):
        if speed<=p:
            return colorList[index]
    else:
        return colorList[0]
def reward2Color(reward,rewardList,colorList):
    if reward > 2:
        return 'red'
    elif reward >= 1:
        return 'orangered'
    else:
        return 'green'
if __name__ == '__main__':
    X,Y = getPathXY()
    plt.axis([-100,max(X)+100, -100,max(Y)+100])
    #plt.ion()
    plt.scatter(X, Y,color='#313D46',linewidth=38)
    fileList = [x for x in os.listdir() if x.lower().endswith('.csv') and x.startswith('0')]
    if len(fileList) == 1:
        dataRow = [x for x in getCSVdata(fileList[0]).split('\n') if x]
        dataConvRow = []
        head = []
        for index,value in enumerate(dataRow):
            if index == 0:
                head = [x.strip() for x in value.replace('action,','').split(',')]
            else:
                dataConvRow.append([x.strip() for x in (value[:value.find('[')] + value[value.find('],')+2:]).split(',')])
        df = pd.DataFrame(data=dataConvRow,columns=head)
        df.episode,df.steps,df.closest_waypoint = df.episode.map(int),df.steps.map(float).map(int),df.closest_waypoint.map(int)
        df.X,df.Y,df.throttle,df.reward,df.progress,df.track_len,df.tstamp = df.X.map(float),df.Y.map(float),df.throttle.map(float),df.reward.map(float),df.progress.map(float),df.track_len.map(float),df.tstamp.map(float)
        dfS = df[df.episode.isin([4])].copy(deep=True)
        ### 1. speed select
        # speedThrottle = 3.0
        # Lx,Ly = [100*x for x in dfS[dfS.throttle < speedThrottle].X.tolist()],[100*y for y in dfS[dfS.throttle < speedThrottle].Y.tolist()]
        # plt.scatter(Lx, Ly,marker = 'o', color = 'green', s = 24)
        ### 2. progress
        progressRange = (15,50)
        Lx,Ly = [100*x for x in dfS[(dfS.progress > progressRange[0]) & (dfS.progress  < progressRange[1])].X.tolist()],[100*y for y in dfS[(dfS.progress > progressRange[0]) & (dfS.progress  < progressRange[1])].Y.tolist()]
        plt.scatter(Lx, Ly,marker = 'o', color = 'green', s = 24)
        plt.show()
    else:
        print('CSV file not unique!')

    

对各个维度进行分析

对速度低于 1.5 的点进行筛选

speedThrottle = 1.5

对赛道范围的进行筛选

progressRange = (15,40)