根据企查查2021-8-18

知乎上的回答
tensorflow官方认证地址
考取google Tensorflow认证的流程
另一个老外的TensorFlow考试经历

Tensorflow考试准备

官方考试手册
配置考试环境
官方教程

学习教程

吴恩达课程学习笔记

AI技术文章

容量、过拟合和欠拟合
深度学习中过拟合、欠拟合问题及解决方案
熵,交叉熵,二分类交叉熵/Entropy, crossentropy, binary crossentropy
正则化
正则化消除过拟合

AI前沿思考

陈德旺对Deepmind的成功的探讨

笔记本电脑上训练

PS: 可以通过Nvidia的显卡加速, 不过在i7-8550U+MX150上测试, MX150似乎比CPU还慢.

使用MX150:

2021-08-23 11:01:05.658828: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cufft64_10.dll
2021-08-23 11:01:05.661719: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library curand64_10.dll
2021-08-23 11:01:05.667080: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cusolver64_11.dll
2021-08-23 11:01:05.672298: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cusparse64_11.dll
2021-08-23 11:01:05.681747: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudnn64_8.dll
2021-08-23 11:01:05.681986: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1871] Adding visible gpu devices: 0
2021-08-23 11:01:05.682349: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX AVX2
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2021-08-23 11:01:05.683329: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1733] Found device 0 with properties: 
pciBusID: 0000:01:00.0 name: NVIDIA GeForce MX150 computeCapability: 6.1
coreClock: 1.341GHz coreCount: 3 deviceMemorySize: 2.00GiB deviceMemoryBandwidth: 37.33GiB/s
2021-08-23 11:01:05.683674: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1871] Adding visible gpu devices: 0
2021-08-23 11:01:06.781669: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1258] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-08-23 11:01:06.781862: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1264]      0 
2021-08-23 11:01:06.781974: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1277] 0:   N 
2021-08-23 11:01:06.784405: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1418] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 1332 MB memory) -> physical GPU (device: 0, name: NVIDIA GeForce MX150, pci bus id: 0000:01:00.0, compute capability: 6.1)
2021-08-23 11:01:07.432975: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:176] None of the MLIR Optimization Passes are enabled (registered 2)
Epoch 1/5
2021-08-23 11:01:07.782500: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cublas64_11.dll
2021-08-23 11:01:08.920483: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cublasLt64_11.dll
1875/1875 [==============================] - 6s 2ms/step - loss: 0.2974 - accuracy: 0.9155
Epoch 2/5
1875/1875 [==============================] - 5s 2ms/step - loss: 0.1430 - accuracy: 0.9569
Epoch 3/5
1875/1875 [==============================] - 5s 2ms/step - loss: 0.1057 - accuracy: 0.9685
Epoch 4/5
1875/1875 [==============================] - 5s 2ms/step - loss: 0.0867 - accuracy: 0.9742
Epoch 5/5
1875/1875 [==============================] - 5s 3ms/step - loss: 0.0741 - accuracy: 0.9764
313/313 - 1s - loss: 0.0704 - accuracy: 0.9784

Process finished with exit code 0

上面MX150的Epoch每一项在5~6秒, 而使用CPU只需要1~2秒:

D:\r\pyproj\TFproj1\venv\Scripts\python.exe D:/r/pyproj/TFproj1/main.py
2021-08-23 11:04:35.810645: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudart64_110.dll
2.5.0
2021-08-23 11:04:39.140247: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library nvcuda.dll
2021-08-23 11:04:39.791839: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1733] Found device 0 with properties: 
pciBusID: 0000:01:00.0 name: NVIDIA GeForce MX150 computeCapability: 6.1
coreClock: 1.341GHz coreCount: 3 deviceMemorySize: 2.00GiB deviceMemoryBandwidth: 37.33GiB/s
2021-08-23 11:04:39.792157: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudart64_110.dll
2021-08-23 11:04:39.808505: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cublas64_11.dll
2021-08-23 11:04:39.808676: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cublasLt64_11.dll
2021-08-23 11:04:39.813804: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cufft64_10.dll
2021-08-23 11:04:39.816603: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library curand64_10.dll
2021-08-23 11:04:39.822109: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cusolver64_11.dll
2021-08-23 11:04:39.827115: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cusparse64_11.dll
2021-08-23 11:04:39.832314: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'cudnn64_8.dll'; dlerror: cudnn64_8.dll not found
2021-08-23 11:04:39.832637: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1766] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
2021-08-23 11:04:39.833777: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX AVX2
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2021-08-23 11:04:39.834660: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1258] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-08-23 11:04:39.834922: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1264]      
2021-08-23 11:04:40.297728: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:176] None of the MLIR Optimization Passes are enabled (registered 2)
Epoch 1/5
1875/1875 [==============================] - 2s 840us/step - loss: 0.2934 - accuracy: 0.9141
Epoch 2/5
1875/1875 [==============================] - 1s 764us/step - loss: 0.1421 - accuracy: 0.9578
Epoch 3/5
1875/1875 [==============================] - 2s 853us/step - loss: 0.1091 - accuracy: 0.9674
Epoch 4/5
1875/1875 [==============================] - 2s 1ms/step - loss: 0.0881 - accuracy: 0.9730
Epoch 5/5
1875/1875 [==============================] - 2s 1ms/step - loss: 0.0763 - accuracy: 0.9759
313/313 - 0s - loss: 0.0729 - accuracy: 0.9780

Process finished with exit code 0

以前认为至少两条：
1、预测准；
2、对人好；
现在得加第三条：
3、执行力强；
要强成信仰一样。

原文连接
启用方式是以管理员运行下面的bat文件:

@echo off
pushd "%~dp0"
dir /b C:\Windows\servicing\Packages\Microsoft-Windows-GroupPolicy-ClientExtensions-Package~3*.mum >List.txt
dir /b C:\Windows\servicing\Packages\Microsoft-Windows-GroupPolicy-ClientTools-Package~3*.mum >>List.txt
for /f %%i in ('findstr /i . List.txt 2^>nul') do dism /online /norestart /add-package:"C:\Windows\servicing\Packages\%%i"
pause

原文来自微软官方

系统文件检查器是 Windows 10 中的一个实用工具，用于检查计算机中文件的问题。若要运行它，请按照下列步骤进行操作：

确保已安装 Windows 10 的最新更新，然后重启计算机。若要了解详细信息，请阅读更新 Windows 10。

在任务栏上的搜索框中，键入“命令提示符”，然后长按或右键单击结果列表内的“命令提示符”（桌面应用）。选择“以管理员身份运行”，然后选择“是”。

键入 DISM.exe /Online /Cleanup-image /Restorehealth（注意每个 "/" 前的空格），然后按“Enter”键。（注意：此步骤可能需要几分钟才会启动，最多可能需要 30 分钟才能完成。）

在看到显示“操作已成功完成”的消息后，键入sfc /scannow（注意“sfc”和“/”之间的空格），然后按下“Enter”。

在看到一条显示“验证 100% 完成”的消息后，键入“退出”。

这么洁身自好竟然中病毒了? 上周五下午去小米换电池, 难道是小米维修?

症状:

1. CPU占用率极高, 一晚上没有关机电脑很烫.
2. 不能打开任务管理器, 提示被管理员禁用.
3. 不能打开组策略管理器 GPEDIT.msc.

腾讯电脑管家居然不报.

C盘 AutoRun.inf

我只能转为图片上传, 因为有的杀毒软件会检测这些病毒内容,并且自动断开与网站的链接.

D 盘 AutoRun.inf

jbqgma.exe

100 KB (103,140 字节)

ulvx.pif

在 https://r.virscan.org/ 上扫描, 结果如下:
文件名称 :ulvx.pif （本站不提供任何文件的下载服务）
文件大小 :103140 byte
文件类型 :PE32 executable (GUI) Intel 80386
MD5:b7ff0959b1568d9530a4c69661dbebc7
SHA1:1b3b4f74fd93e1786c2c195125c28aaae5d8f57b
SHA256:9e9d4b2eaf3c20b3c4397c205989970d2064a7d1a8ced715e62c955d3f8922e5
SSDEEP:1536:5ww2cwmgypHD4U4EYtBnp9LFT8RkTWvaYGTQ2fACZWoWafNRyX2QCSQeAxkR5DlY:5zDJD4CYzxLCig2aoWPmQCSQlkR5m

■Heuristic/Suspicious ■Exact
注意: 就算报告发现病毒，也可能是杀软误报，请根据查毒结果自行判断

分析

应该是在最近中的病毒. 一着急没去看病毒文件的创建日期就给删掉了. 看起来似乎是U盘病毒. 除了在小米换电池, 还有就是上周去出差电脑拿出来做演示, 是不是有病毒文件通过别人的U盘传入? 就不知道了.

杀毒

使用Avast全盘杀毒, 查出来3660个病毒, 大多是Win32: SaliCode[Inf]病毒,还有一些是Wrm病毒, 感染的主要是Exe文件. 我使用everything搜索所有Exe文件并对修改时间排序, 除了Avast正在查杀修改的Exe以外, 竟然找不到其他大批量修改Exe的时间段?!

太极创客对28BYJ-48单极性步进电机的介绍很详细, 型号源于名称: 外径28毫米四相八拍式永磁减速型步进电机.B是步进,Y是永磁, J是减速,哈哈很明显是我国的电机标准.
太极创客对这个步进电机的原理的介绍讲转子有6齿而定子有8齿, 最后算下来每一步是11.25度, 很有道理但很可惜, 是错的. 按转子6齿, BCDA每次循环完毕, 转子只会转45/3×4=60度, 每一步只有45/3=15度. 但11.25度这个结论却是对的. 是因为其实转子和定子都是8齿!
这个老外的博客对28BYJ-48这个国产型号电机进行了详细的拆解分析, 反而是正确的.
定子是4个8齿的齿板组成, 分上下两组. 每组两个齿板两两相对. 每个齿板上, 齿角是360/8=45度. 从上到下四个齿板角度依次为0度, (45/4)×2度, (45/4)×1度, (45/4)×3度. 转子是8个方向的永磁铁组成. 每一步可以走45/4=11.25度. 走完一圈是360/11.25=32步. 减速比1/64, 故最外层走完一圈是32×64=2048步. 然而上下两组齿板同时通磁,可以形成半步, 这样可以做到2048×2=4096个不同角度.

直观上, 舵机的噪音好大.
我买的这款步进电机28BYJ-48, 在转动的时候似乎在不停的震颤. 难道是步进的缘故?

Arduino基础入门篇28—舵机控制

NodeMCU需要注意的是: D0~D8并不是对应GPIO0~8,而在Arduino中需要使用的PIN是GPIO的编号.
对应关系是:

文章A
1.0/0.9/V3版针脚图(PDF)
0.9版原理图(JPG)

原文链接: ESP8266 NodeMCU – Controling Stepper Motor with ULN2003 Driver

最主要是2个:

1. 使用了AccelStepper库,而不是常见的Stepper库.
2. 电机要单独供电. 文章使用9V+LM7805稳压到5V, 我就直接用直流电源了. 5.3V下电流大约0.17A.

程序如下:

#include <AccelStepper.h>           // Load the AccelStepper library

#define motorPin1  5                // IN1 pin on the ULN2003A driver to pin D1 on NodeMCU board
#define motorPin2  4                // IN2 pin on the ULN2003A driver to pin D2 on NodeMCU board
#define motorPin3  0                // IN3 pin on the ULN2003A driver to pin D3 on NodeMCU board
#define motorPin4  2                // IN4 pin on the ULN2003A driver to pin D4 on NodeMCU board

int stepsPerRevolution = 64;        // steps per revolution
int degreePerRevolution = 5.625;    // degree per revolution

/*
 * AccelStepper::FULL2WIRE (2) means: 2 wire stepper (2 pins needed). 
 * AccelStepper::FULL3WIRE (3) means: 3 wire stepper, like a harddisk motor (3 pins needed). 
 * AccelStepper::FULL4WIRE (4) means: 4 wire stepper (4 pins needed). 
 * AccelStepper::HALF3WIRE (6) means: 3 wire half stepper, like a harddisk motor (3 pins needed) 
 * AccelStepper::HALF4WIRE (8) means: 4 wire half stepper (4 pins needed) 
 *
 * AccelStepper uses AccelStepper::FULL4WIRE (4 pins needed) by default.
 */
AccelStepper stepper(AccelStepper::HALF4WIRE, motorPin1, motorPin3, motorPin2, motorPin4);

void setup() {
  Serial.begin(9600);               // initialise the serial monitor

  stepper.setMaxSpeed(1000.0);      // set the max motor speed
  stepper.setAcceleration(100.0);   // set the acceleration
  stepper.setSpeed(200);            // set the current speed

}

// Degrees of the momevement. So first to 90 degrees, then to -90 graden then to 180 degrees etc.
int steps[] = { 90, -90, 180, -180, 0 };

// The total entries in steps[]
int stepsCount = 5;

// Keeps track of the position in steps[] we are about to run
int stepsIndex = 0;

void loop() {

  // If the stepper isn't moving and doesn't have to go any distance
  if (!stepper.isRunning() && stepper.distanceToGo() == 0) {

    // Move the stepper to the degrees on the position of stepIndex in steps[]
    stepper.moveTo(degToSteps(steps[stepsIndex]));

    // Increase the index when each movement has finished
    stepsIndex++;

    // If we have executed all positions of steps[]
    if (stepsIndex > stepsCount) {

      // Set the index to 0 to restart the process
      stepsIndex = 0;
    }
  }

  stepper.run();

}

/*
 * Converts degrees to steps
 * 
 * 28BYJ-48 motor has 5.625 degrees per step
 * 360 degrees / 5.625 = 64 steps per revolution
 * 
 * Example with degToSteps(45):
 * (64 / 5.625) * 45 = 512 steps
 */
float degToSteps(float deg) {
  return (stepsPerRevolution / degreePerRevolution) * deg;
}

这个程序有个问题是, 断电后重新上电, 是以当前位置为0位的, 不能记忆上次断电的位置.

空调中的导风口使用的也是步进电机, 这儿有一篇维修文章

参考文章:
AccelStepper官方文档
28BYJ-48 Stepper Motor with ULN2003 Driver and Arduino Tutorial
ULN2003-Datasheet.pdf
ULN2003-Stepper-Motor-Driver-PCB.pdf
28byj48-Stepper-Motor-Datasheet.pdf

网易云音乐上的翻译完全失去了原词的性感, 这儿我对比重新翻译一下。

作词 : Adam Levine/Shellback/Savan Kotecha/Max Martin
作曲 : Adam Levine/Shellback/Savan Kotecha/Max Martin

官方的说明上不去了 http://docs.typecho.org/develop/acl， 这儿转载官方说明如下：

在Typecho中分别设置了5个等级的权限供多个用户共用一个博客。他们的权限从高到低分别是:管理员(administrator), 编辑(editor), 贡献者(contributor), 关注者(subscriber), 访问者(visitor).其权限设置与wordpress有一些相同之处，但是扩展方法不尽相同。

开发人员接口

对用户的权限操作以及用户信息获取，我们都封装在了Widget_User组件中。开发人员可以通过调用此组件获取相关信息。

1. 数字越小权限越高
2. 审核后发布
3. , 5) 仅限于自己
4. 受限

只知非黑即白的年轻人喜欢说话，他们觉得世界和自己一样单纯；只知流量为王的自媒体喜欢说话，他们觉得世界处处都能把流量变现；剩下的我们知道世界善恶相杂、并不易分开对错，所以不爱说话，所以，我们是沉默的大多数。😷

知乎：有哪些好用的管理家庭物品的应用（收纳管理）？

风险

• 法律风险： @晒书房 在2021-7-23发布微博称将于2021-7-30停止服务，出于个人兴趣至今已经服务9年多，微博粉丝40167，看起来似乎是监管所致。物品、书籍拥有即有合法非法之分
• 展示风险： 物品近乎于个人隐私。可以从物品的拥有判断出此人的兴趣爱好等。故不宜将所有物品直接展示。
• 交易风险： 如果有人借助此平台进行非法交易，如贩卖毒品该怎么办？交易中涉及钱财交换，如何相互信任？所以不宜通过此平台直接交易，而宜导流至其他成熟平台如闲鱼。但这样做会大大增加用户交易付出的精力和时间成本，导致交易流动性减弱。