(进阶)AWS DeepRacer获取log-001

痛定思痛。 2022-08-28 06:59 9阅读 0赞

### 文章目录 ###

*  如何获取log
 *  一个打印log的代码示例
 *  下载log
 *  training log 文件结构
 *   *  自定义log位置
     *  自定义log内容
 *  Evaluation log文件结构

# 如何获取log #

在`reward_function` 中通过print可以将params入口参数打印到log中，然后通过控制台Download logs可以获取自定义的log格式  
![在这里插入图片描述][watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBAUmFtYm8uRmFu_size_20_color_FFFFFF_t_70_g_se_x_16]

# 一个打印log的代码示例 #

def reward_function(params):
        ''' Env: re:Invent 2018 '''
        x = params['x']
        y = params['y']
        next_waypoints=params['closest_waypoints'][1]
        is_crashed=params['is_crashed']
        is_left_of_center=params['is_left_of_center']
        is_offtrack=params['is_offtrack']
        is_reversed=params['is_reversed']
        heading=params['heading']
        progress=params['progress']
        steering_angle=params['steering_angle']
        steps=params['steps']
        track_length=params['track_length']
        waypoints=params['waypoints']
        track_width = params['track_width']
        distance_from_center = params['distance_from_center']
        all_wheels_on_track = params['all_wheels_on_track']
        speed = params['speed']
        abs_steering = abs(params['steering_angle'])
        reward = 1.0
        
        def printLog():
            print('reward_constant_Log,{track_length},{waypoints}'.format(track_length=track_length,waypoints=waypoints))
            print('''reward_var_Log,{steps},{next_waypoints},{x},{y},{all_wheels_on_track},{speed},{reward},{distance_from_center},{progress},{steering_angle},{heading},{track_width},{is_offtrack},{is_crashed},{is_left_of_center},{is_reversed}'''.format(
            steps=steps,
            next_waypoints=next_waypoints,
            x=x,
            y=y,
            all_wheels_on_track=all_wheels_on_track,
            speed=speed,
            reward=reward,
            distance_from_center=distance_from_center,
            progress=progress,
            steering_angle=steering_angle,
            heading=heading,
            track_width=track_width,
            is_offtrack=is_offtrack,
            is_crashed=is_crashed,
            is_left_of_center=is_left_of_center,
            is_reversed=is_reversed
            ))
    
        printLog()
        
        return float(reward)

# 下载log #

下载下来的log 是

'{modelName}-training_job_xxxx_logs.tar.gz'

![在这里插入图片描述][efbfa34461a4458c85d96a27a2d37025.png]

# training log 文件结构 #

以下是解压后log文件结构，以及每个文件夹下文件记录的数据的来源，此处只做最简单的介绍后期日志分析时再做进一步深入探究。

│  2018-getlog-training_job_SS2xzf4LQgmyiUdiaAhPNw_logs.tar.gz
    │  GetLog.py
    │  getLogExample.py
    │
    └─29e9138b-9fd9-4f01-8952-a4d4dd7a02b8
        ├─logs                                        # logs下为训练过程日志
        │  └─training
        │          training-20211009172232-SS2xzf4LQgmyiUdiaAhPNw-sagemaker.log
        │          training-20211009172233-SS2xzf4LQgmyiUdiaAhPNw-robomaker.log     # 自定义的日志将放在这里
        │
        ├─metrics                              # metrics下为训练过程指标
        │  └─training
        │          training-20211009164204-SS2xzf4LQgmyiUdiaAhPNw.json
        │
        └─sim-trace                         # sim-trace下为每个iteration下的trace以及对应reward的记录
            └─training
                └─training-simtrace
                        0-iteration.csv
                        1-iteration.csv
                        10-iteration.csv
                        11-iteration.csv
                        12-iteration.csv
                        2-iteration.csv
                        3-iteration.csv
                        4-iteration.csv
                        5-iteration.csv
                        6-iteration.csv
                        7-iteration.csv
                        8-iteration.csv
                        9-iteration.csv

## 自定义log位置 ##

![在这里插入图片描述][051df1b8f92343e3bf9a9a42f8007fae.png]

## 自定义log内容 ##

下一篇文章将会对这个log提取之后进行深入分析  
![在这里插入图片描述][watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBAUmFtYm8uRmFu_size_20_color_FFFFFF_t_70_g_se_x_16 1]

# Evaluation log文件结构 #

![在这里插入图片描述][watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBAUmFtYm8uRmFu_size_20_color_FFFFFF_t_70_g_se_x_16 2]

[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBAUmFtYm8uRmFu_size_20_color_FFFFFF_t_70_g_se_x_16]: /images/20220828/566e1851c30b48a38789b2140269e006.png
[efbfa34461a4458c85d96a27a2d37025.png]: /images/20220828/6ea9ed3fc4df43338cdd712b5f39a07b.png
[051df1b8f92343e3bf9a9a42f8007fae.png]: /images/20220828/96a86238d9d4441289e863b0ec73c230.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBAUmFtYm8uRmFu_size_20_color_FFFFFF_t_70_g_se_x_16 1]: /images/20220828/e2ac60d09848493db91cf47c55ca92d6.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBAUmFtYm8uRmFu_size_20_color_FFFFFF_t_70_g_se_x_16 2]: /images/20220828/3575ee12b6c14eeeb3679cecacf1f262.png