DL之CharRNN：人工智能为你写英语小说短文/中文写诗/中文写歌词/写代码——基于TF利用RNN算法训练数据集(比如Coriolanus等)、训练&测试过程全记录

conce alone,
Which treason thines, and true a mercy with the man,
And honour the sheet in my stal and taste and him,
I house your servants there, and a shall some
They take a thing a bate of him, but then, but he is were
Too hath to the match her wanders there he wants.

APRINS:
The sender, and the that's all him at thy bloods;
To thee an ast that that the solet is a wealer this thar was a
will seemen of the than a court to me were burist.

LUCIANO:
I am how, an are this holds and to her sad
Woo on your toon and as thing, have the will
A show that's tears of my satate, some are.

SIR TOBY RELLA:
If the word of her of this sad of my mans are?

ALINA:
I should say thou the thee with an erms, we'll he is with his better.
ANTELIUS:
Will you have those stalling and as a wild
In that, have would say well, he will be string that all her
to here, that a bosom and at as my bands to--they
women is shall say this wears thee;
Hath see, a say me off the honour to the with
That wither and sting, and

独自一人，
哪一个叛逆变瘦，真正的仁慈，
并在我的石板上品尝这张纸，品尝他，
我把你的仆人安放在那里
他们把一件事拿给他听，但他现在是
她想去的地方，也有她的对手。
APRINS：
发送者，那就是他在你的血液里；
对你来说，一个让自己成为一个富人的圣地
对我来说似乎比法庭更重要。
LUCIANO：
我是如何，这是持有和她的悲伤
对你的香椿和作为事物，有意志
这是我的眼泪，有些是。
SIR TOBY RELLA：
如果她的话我的男人的这种悲伤是什么？
艾琳娜：
我应该说你是一个厄尔默斯，我们会和他更好相处。
ANTELIUS：
你会有那些失速和野性吗？
在那，有人会说，他会把她所有的绳子
到这里，那一个胸怀和我的乐队
女人应该说这穿了你；
看见了，向我告别
枯萎和刺痛，

2、test02

fort,
To hear mine aldouse, and hath have a tongue of her as mine.

SILVIA:
And so and merty to this hand,
To seal you we would take the thought in a time.

POLANIA:
I am all their mind, that this the tongue.

ANDELIO:
He thither at my badis, tink as thou
handly; and and all thy last at tell thee
Tell you sake a court of thou to my store with him of my shame
To any shild a shild and heaven as her throog
As his meant and hath to my man and sence,
I had all the witere the storal one worth.

PRINCE HENRY:
He hath something to stay.

SIR OF ORIANIO:
He have troubles here.

LAUVANICUS:
With made that the stally that to this horses, and terter.

PALIS:
Is have this true any stone to the with the sold,
I were a master, the tongue, that a sharlon on
me to he saw and may thy stallangers to streak,
The weads to the more so fiers in thee. And you think
What is, to honour to thy house at
their stays and hand, whell string, thyness would have this,
A soul to-news would be the his hands.
He love you

堡垒，
听我的话，她有我的舌头。
SILVIA：
所以，梅蒂，这只手，
为了封你，我们会在一段时间内接受这个想法。
POLANIA：
我都是他们的头脑，这就是舌头。
ANDELIO：
他在我的坏蛋那里，丁克如你
你最后一次告诉你
你把我的羞辱告诉你我的商店
对任何一个阴霾和天堂作为她的悸动
正如他对我的男人和我所拥有的，
我拥有所有值得收藏的东西。
PRINCE HENRY：
他有事要留下。
奥里亚尼奥爵士：
他这里有麻烦。
LAUVANICUS：
就这样把那匹马吓了一跳。
PALIS：
是真的有什么石头卖给谁，
我是一个大师，舌头，一个沙龙
我见他，愿你的斯塔兰格连任，
在你身上，维斯的音符越来越高。你认为
什么是荣耀你的家
他们的停留和手，弦，thyess都有这个，
新闻的灵魂是他的手。
他爱你

监控模型

训练过程全记录

2018-10-13 17:05:49.402137: 
step: 10/20000...  loss: 3.4659...  0.1860 sec/batch

……

step: 1000/20000...  loss: 2.0612...  0.1168 sec/batch

……

step: 2000/20000...  loss: 1.9092...  0.1278 sec/batch

……

step: 3000/20000...  loss: 1.8643...  0.1283 sec/batch

……

step: 10000/20000...  loss: 1.8001...  0.1329 sec/batch

……

step: 15000/20000...  loss: 1.7402...  0.1689 sec/batch
step: 15010/20000...  loss: 1.8033...  0.2306 sec/batch
step: 15020/20000...  loss: 1.8284...  0.1499 sec/batch
step: 15030/20000...  loss: 1.7952...  0.1359 sec/batch
step: 15040/20000...  loss: 1.7906...  0.1514 sec/batch
step: 15050/20000...  loss: 1.7777...  0.1053 sec/batch
step: 15060/20000...  loss: 1.7665...  0.1298 sec/batch
step: 15070/20000...  loss: 1.7931...  0.1183 sec/batch
step: 15080/20000...  loss: 1.8027...  0.1404 sec/batch
step: 15090/20000...  loss: 1.8116...  0.1238 sec/batch
step: 15100/20000...  loss: 1.7969...  0.1108 sec/batch

……

step: 19800/20000...  loss: 1.8298...  0.1233 sec/batch
step: 19810/20000...  loss: 1.8231...  0.1228 sec/batch
step: 19820/20000...  loss: 1.7674...  0.1329 sec/batch
step: 19830/20000...  loss: 1.7872...  0.1434 sec/batch
step: 19840/20000...  loss: 1.8333...  0.1228 sec/batch
step: 19850/20000...  loss: 1.6446...  0.1464 sec/batch
step: 19860/20000...  loss: 1.8021...  0.1509 sec/batch
step: 19870/20000...  loss: 1.8217...  0.1168 sec/batch
step: 19880/20000...  loss: 1.7298...  0.1178 sec/batch
step: 19890/20000...  loss: 1.6948...  0.1293 sec/batch
step: 19900/20000...  loss: 1.7582...  0.1253 sec/batch
step: 19910/20000...  loss: 1.8246...  0.1414 sec/batch
step: 19920/20000...  loss: 1.7258...  0.1103 sec/batch
step: 19930/20000...  loss: 1.8216...  0.1544 sec/batch
step: 19940/20000...  loss: 1.7866...  0.1243 sec/batch
step: 19950/20000...  loss: 1.7673...  0.1088 sec/batch
step: 19960/20000...  loss: 1.7285...  0.1088 sec/batch
step: 19970/20000...  loss: 1.7658...  0.1073 sec/batch
step: 19980/20000...  loss: 1.8054...  0.1198 sec/batch
step: 19990/20000...  loss: 1.7714...  0.1128 sec/batch
step: 20000/20000...  loss: 1.7530...  0.1228 sec/batch

训练的数据集

《科利奥兰纳斯》是莎士比亚晚年撰写的一部罗马历史悲剧，讲述了罗马共和国的英雄马歇斯（被称为科利奥兰纳斯），因性格多疑、脾气暴躁，得罪了公众而被逐出罗马的悲剧。作者以英雄与群众的关系为主线，揭示出人性的弱点。

1、部分章节

First Citizen:
Before we proceed any further, hear me speak.

All:
Speak, speak.

First Citizen:
You are all resolved rather to die than to famish?

All:
Resolved. resolved.

First Citizen:
First, you know Caius Marcius is chief enemy to the people.

All:
We know't, we know't.

First Citizen:
Let us kill him, and we'll have corn at our own price.
Is't a verdict?

All:
No more talking on't; let it be done: away, away!

Second Citizen:
One word, good citizens.

First Citizen:
We are accounted poor citizens, the patricians good.
What authority surfeits on would relieve us: if they
would yield us but the superfluity, while it were
wholesome, we might guess they relieved us humanely;
but they think we are too dear: the leanness that
afflicts us, the object of our misery, is as an
inventory to particularise their abundance; our
sufferance is a gain to them Let us revenge this with
our pikes, ere we become rakes: for the gods know I
speak this in hunger for bread, not in thirst for revenge.

Second Citizen:
Would you proceed especially against Caius Marcius?

All:
Against him first: he's a very dog to the commonalty.

Second Citizen:
Consider you what services he has done for his country?

First Citizen:
Very well; and could be content to give him good
report fort, but that he pays himself with being proud.

Second Citizen:
Nay, but speak not maliciously.

First Citizen:
I say unto you, what he hath done famously, he did
it to that end: though soft-conscienced men can be
content to say it was for his country he did it to
please his mother and to be partly proud; which he
is, even till the altitude of his virtue.

Second Citizen:
What he cannot help in his nature, you account a
vice in him. You must in no way say he is covetous.

First Citizen:
If I must not, I need not be barren of accusations;
he hath faults, with surplus, to tire in repetition.
What shouts are these? The other side o' the city
is risen: why stay we prating here? to the Capitol!

All:
Come, come.

First Citizen:
Soft! who comes here?

Second Citizen:
Worthy Menenius Agrippa; one that hath always loved
the people.

First Citizen:
He's one honest enough: would all the rest were so!

人工智能为你写诗——基于TF利用RNN算法实现【机器为你写诗】、训练&测试过程全记录

输出结果

1、test01

<unk><unk>风下，天上不相逢。
一人不得别，不得不可寻。
何事无时事，谁知一日年。
一年多旧国，一处不相寻。
白发何人见，清辉自有人。
何当不知事，相送不相思。
此去无相见，何年有旧年。
相思一阳处，相送不能归。
不得千株下，何时不得时。
此心不可见，此日不相思。
何事有归客，何言不见时。
何处不知处，不知山上时。
何当不可问，一事不知情。
此去不知事，无人不有身。
相逢何处见，不觉白云间。
白云一相见，一月不知时。
一夜一秋色，青楼不可知。
江西山水下，山下白头心。
此事何人见，孤舟向日深。
何当无一事，此路不相亲。
何必有山里，不知归去人。
相知不相识，此地不能闻。
白发千株树，秋风落日寒。

2、test02

不得无时人，何人不知别。
不见一时人，相思何日在。
一夜不可知，何时无一日。
一人不自见，何处有人情。
何处不知此，何人见此年。
何时不得处，此日不知君。
一日何人得，何由有此身。
相思无不得，一去不知君。
此日何人见，东风满月中。
无知不得去，不见不相思。
不得千峰上，相逢独不穷。
何年不可待，一夜有春风。
不见南江路，相思一夜归。
何人知此路，此日更难归。
白日无人处，秋光入水流。
江南无限路，相忆在山城。
何处知何事，孤城不可归。
山边秋月尽，江水水风生。
不有南江客，何时有故乡。
山中山上水，山上水中风。
何处不相访，何年有此心。
江边秋雨尽，山水白云寒。

3、test03

一朝多不见，无处在人人。
一里无人去，无人见故山。
何年不知别，此地有相亲。
白首何时去，春风不得行。
不知归去去，谁见此来情。
不有青山去，谁知不自知。
一时何处去，此去有君人。
不见南江路，何时是白云。
不知归客去，相忆不成秋。
不得东西去，何人有一身。
不能知此事，不觉是人情。
一里无时事，相逢不得心。
江南秋雨尽，风落夜来深。
何处不知别，春来不自知。
一朝无一处，何处是江南。
一里不相见，东西无一人。
一朝归去处，何事见沧洲。
此去无年处，何年有远心。
不知山上客，不是故人心。
一里不可得，无人有一年。
何人不见处，不得一朝生。
不见青云客，何人是此时。

模型监控

训练、测试过程全记录

1、训练过程



2018-10-13 22:31:33.935213:
step: 10/10000...  loss: 6.5868...  0.1901 sec/batch
step: 20/10000...  loss: 6.4824...  0.2401 sec/batch
step: 30/10000...  loss: 6.3176...  0.2401 sec/batch
step: 40/10000...  loss: 6.2126...  0.2401 sec/batch
step: 50/10000...  loss: 6.0081...  0.2301 sec/batch
step: 60/10000...  loss: 5.7657...  0.2401 sec/batch
step: 70/10000...  loss: 5.6694...  0.2301 sec/batch
step: 80/10000...  loss: 5.6661...  0.2301 sec/batch
step: 90/10000...  loss: 5.6736...  0.2301 sec/batch
step: 100/10000...  loss: 5.5698...  0.2201 sec/batch
step: 110/10000...  loss: 5.6083...  0.2301 sec/batch
step: 120/10000...  loss: 5.5252...  0.2301 sec/batch
step: 130/10000...  loss: 5.4708...  0.2301 sec/batch
step: 140/10000...  loss: 5.4311...  0.2401 sec/batch
step: 150/10000...  loss: 5.4571...  0.2701 sec/batch
step: 160/10000...  loss: 5.5295...  0.2520 sec/batch
step: 170/10000...  loss: 5.4211...  0.2401 sec/batch
step: 180/10000...  loss: 5.4161...  0.2802 sec/batch
step: 190/10000...  loss: 5.4389...  0.4943 sec/batch
step: 200/10000...  loss: 5.3380...  0.4332 sec/batch

……
step: 790/10000...  loss: 5.2031...  0.2301 sec/batch
step: 800/10000...  loss: 5.2537...  0.2301 sec/batch
step: 810/10000...  loss: 5.0923...  0.2301 sec/batch
step: 820/10000...  loss: 5.1930...  0.2501 sec/batch
step: 830/10000...  loss: 5.1636...  0.2535 sec/batch
step: 840/10000...  loss: 5.1357...  0.2801 sec/batch
step: 850/10000...  loss: 5.0844...  0.2236 sec/batch
step: 860/10000...  loss: 5.2004...  0.2386 sec/batch
step: 870/10000...  loss: 5.1894...  0.2401 sec/batch
step: 880/10000...  loss: 5.1631...  0.2501 sec/batch
step: 890/10000...  loss: 5.1297...  0.2477 sec/batch
step: 900/10000...  loss: 5.1044...  0.2401 sec/batch
step: 910/10000...  loss: 5.0738...  0.2382 sec/batch
step: 920/10000...  loss: 5.0971...  0.2701 sec/batch
step: 930/10000...  loss: 5.1829...  0.2501 sec/batch
step: 940/10000...  loss: 5.1822...  0.2364 sec/batch
step: 950/10000...  loss: 5.1883...  0.2401 sec/batch
step: 960/10000...  loss: 5.0521...  0.2301 sec/batch
step: 970/10000...  loss: 5.0848...  0.2201 sec/batch
step: 980/10000...  loss: 5.0598...  0.2201 sec/batch
step: 990/10000...  loss: 5.0421...  0.2701 sec/batch
step: 1000/10000...  loss: 5.1234...  0.2323 sec/batch
step: 1010/10000...  loss: 5.0744...  0.2356 sec/batch
step: 1020/10000...  loss: 5.0408...  0.2401 sec/batch
step: 1030/10000...  loss: 5.1138...  0.2417 sec/batch
step: 1040/10000...  loss: 4.9961...  0.2601 sec/batch
step: 1050/10000...  loss: 4.9691...  0.2401 sec/batch
step: 1060/10000...  loss: 4.9938...  0.2601 sec/batch
step: 1070/10000...  loss: 4.9778...  0.2601 sec/batch
step: 1080/10000...  loss: 5.0157...  0.3600 sec/batch

……

step: 4440/10000...  loss: 4.7136...  0.2757 sec/batch
step: 4450/10000...  loss: 4.5896...  0.2767 sec/batch
step: 4460/10000...  loss: 4.6408...  0.3088 sec/batch
step: 4470/10000...  loss: 4.6901...  0.2737 sec/batch
step: 4480/10000...  loss: 4.5717...  0.2707 sec/batch
step: 4490/10000...  loss: 4.7523...  0.2838 sec/batch
step: 4500/10000...  loss: 4.7592...  0.2998 sec/batch
step: 4510/10000...  loss: 4.6054...  0.2878 sec/batch
step: 4520/10000...  loss: 4.7039...  0.2868 sec/batch
step: 4530/10000...  loss: 4.6380...  0.2838 sec/batch
step: 4540/10000...  loss: 4.5378...  0.2777 sec/batch
step: 4550/10000...  loss: 4.7376...  0.3098 sec/batch
step: 4560/10000...  loss: 4.7103...  0.2797 sec/batch
step: 4570/10000...  loss: 4.7170...  0.2767 sec/batch
step: 4580/10000...  loss: 4.7307...  0.3168 sec/batch
step: 4590/10000...  loss: 4.7139...  0.3108 sec/batch
step: 4600/10000...  loss: 4.7419...  0.3028 sec/batch
step: 4610/10000...  loss: 4.7980...  0.2918 sec/batch
step: 4620/10000...  loss: 4.7127...  0.2797 sec/batch
step: 4630/10000...  loss: 4.7264...  0.2858 sec/batch
step: 4640/10000...  loss: 4.6384...  0.3419 sec/batch
step: 4650/10000...  loss: 4.6761...  0.2978 sec/batch
step: 4660/10000...  loss: 4.8158...  0.3590 sec/batch
step: 4670/10000...  loss: 4.7579...  0.2828 sec/batch
step: 4680/10000...  loss: 4.7702...  0.2777 sec/batch
step: 4690/10000...  loss: 4.6909...  0.2607 sec/batch
step: 4700/10000...  loss: 4.6037...  0.2808 sec/batch
step: 4710/10000...  loss: 4.6775...  0.2848 sec/batch
step: 4720/10000...  loss: 4.6074...  0.2838 sec/batch
step: 4730/10000...  loss: 4.7280...  0.3088 sec/batch
step: 4740/10000...  loss: 4.7241...  0.3539 sec/batch
step: 4750/10000...  loss: 4.5496...  0.2948 sec/batch
step: 4760/10000...  loss: 4.6488...  0.3189 sec/batch
step: 4770/10000...  loss: 4.6698...  0.3048 sec/batch
step: 4780/10000...  loss: 4.6410...  0.3068 sec/batch
step: 4790/10000...  loss: 4.7408...  0.3329 sec/batch
step: 4800/10000...  loss: 4.6425...  0.2928 sec/batch
step: 4810/10000...  loss: 4.6900...  0.2978 sec/batch
step: 4820/10000...  loss: 4.5715...  0.3499 sec/batch
step: 4830/10000...  loss: 4.7289...  0.2868 sec/batch
step: 4840/10000...  loss: 4.7500...  0.2998 sec/batch
step: 4850/10000...  loss: 4.7674...  0.2968 sec/batch
step: 4860/10000...  loss: 4.6832...  0.3078 sec/batch
step: 4870/10000...  loss: 4.7478...  0.3008 sec/batch
step: 4880/10000...  loss: 4.7895...  0.2817 sec/batch

……

step: 9710/10000...  loss: 4.5839...  0.2601 sec/batch
step: 9720/10000...  loss: 4.4666...  0.2401 sec/batch
step: 9730/10000...  loss: 4.6392...  0.2201 sec/batch
step: 9740/10000...  loss: 4.5415...  0.2201 sec/batch
step: 9750/10000...  loss: 4.6513...  0.2201 sec/batch
step: 9760/10000...  loss: 4.6485...  0.2201 sec/batch
step: 9770/10000...  loss: 4.5317...  0.2201 sec/batch
step: 9780/10000...  loss: 4.5547...  0.2301 sec/batch
step: 9790/10000...  loss: 4.3995...  0.2301 sec/batch
step: 9800/10000...  loss: 4.5596...  0.2301 sec/batch
step: 9810/10000...  loss: 4.5636...  0.2301 sec/batch
step: 9820/10000...  loss: 4.4348...  0.2201 sec/batch
step: 9830/10000...  loss: 4.5268...  0.2201 sec/batch
step: 9840/10000...  loss: 4.5790...  0.2201 sec/batch
step: 9850/10000...  loss: 4.6265...  0.2301 sec/batch
step: 9860/10000...  loss: 4.6017...  0.2401 sec/batch
step: 9870/10000...  loss: 4.4009...  0.2301 sec/batch
step: 9880/10000...  loss: 4.4448...  0.2201 sec/batch
step: 9890/10000...  loss: 4.5858...  0.2201 sec/batch
step: 9900/10000...  loss: 4.5622...  0.2201 sec/batch
step: 9910/10000...  loss: 4.4015...  0.2301 sec/batch
step: 9920/10000...  loss: 4.5220...  0.2301 sec/batch
step: 9930/10000...  loss: 4.5207...  0.2201 sec/batch
step: 9940/10000...  loss: 4.4752...  0.2201 sec/batch
step: 9950/10000...  loss: 4.4572...  0.2301 sec/batch
step: 9960/10000...  loss: 4.5389...  0.2201 sec/batch
step: 9970/10000...  loss: 4.5561...  0.2301 sec/batch
step: 9980/10000...  loss: 4.4487...  0.2401 sec/batch
step: 9990/10000...  loss: 4.4851...  0.2301 sec/batch
step: 10000/10000...  loss: 4.5944...  0.2201 sec/batch

2、测试过程

训练的数据集

1、大量的五言唐诗

寒随穷律变，春逐鸟声开。
初风飘带柳，晚雪间花梅。
碧林青旧竹，绿沼翠新苔。
芝田初雁去，绮树巧莺来。
晚霞聊自怡，初晴弥可喜。
日晃百花色，风动千林翠。
池鱼跃不同，园鸟声还异。
寄言博通者，知予物外志。
一朝春夏改，隔夜鸟花迁。
阴阳深浅叶，晓夕重轻烟。
哢莺犹响殿，横丝正网天。
珮高兰影接，绶细草纹连。
碧鳞惊棹侧，玄燕舞檐前。
何必汾阳处，始复有山泉。
夏律昨留灰，秋箭今移晷。
峨嵋岫初出，洞庭波渐起。
桂白发幽岩，菊黄开灞涘。
运流方可叹，含毫属微理。
寒惊蓟门叶，秋发小山枝。
松阴背日转，竹影避风移。
提壶菊花岸，高兴芙蓉池。
欲知凉气早，巢空燕不窥。
山亭秋色满，岩牖凉风度。
疏兰尚染烟，残菊犹承露。
古石衣新苔，新巢封古树。
历览情无极，咫尺轮光暮。
慨然抚长剑，济世岂邀名。
星旗纷电举，日羽肃天行。
遍野屯万骑，临原驻五营。
登山麾武节，背水纵神兵。
在昔戎戈动，今来宇宙平。
翠野驻戎轩，卢龙转征旆。
遥山丽如绮，长流萦似带。
海气百重楼，岩松千丈盖。
兹焉可游赏，何必襄城外。
玄兔月初明，澄辉照辽碣。
映云光暂隐，隔树花如缀。
魄满桂枝圆，轮亏镜彩缺。
临城却影散，带晕重围结。
驻跸俯九都，停观妖氛灭。
碧原开雾隰，绮岭峻霞城。
烟峰高下翠，日浪浅深明。
斑红妆蕊树，圆青压溜荆。
迹岩劳傅想，窥野访莘情。
巨川何以济，舟楫伫时英。
春蒐驰骏骨，总辔俯长河。
霞处流萦锦，风前漾卷罗。
水花翻照树，堤兰倒插波。
岂必汾阴曲，秋云发棹歌。
重峦俯渭水，碧嶂插遥天。

……

人工智能为你写周董歌词——基于TF利用RNN算法实现【机器为你作词】、训练&测试过程全记录

输出结果

1、test01

夕海
而我在等待之光
在月前被画面
而我心碎 你的个世纪
你的时间
我在赶过去
我的不是你不会感觉妈妈
我说不要不要说 我会爱你
我不要你不会
我不能不会别人
这样的人们我们不要
我会不要再见
你的时光机
这些爱 不要会会爱你
你也能再不要承诺
我只要这种种多简单
你的我的大人 我不会别以
我的感觉不好我走
你说你已经不能够继续
你说你爱我
你不会再想你
不会怕你 我会不要我
我不想再有一口
我要想你我不要再人
包容 你没有回忆
不要我不会感觉不来
我的不要你没有错觉
不能承受我已无奈单
不能要不要再要我会多爱我
能是我不是我
我要离开

2、test01

我说的你爱我的手
我们不能够沉默
我们的感觉
我们的爱情 我们的不是 一个人
让我们乘着阳光
让你在窗窗外面
我一路向北
一直在秋天
你的世界 是我心中不来
你的我爱你 我们
你的爱我 让我给你的美　
我知道这里说你不是我的人爱
这样不会停留我的伤的
我说不要再要 我要一种悲哀
不是你的我不想我不能说说
没有你烦恼
我只能听过
我说我会不要你的爱
你说你不想听你
我不会感动的天  你要的话
我不要我的爱我
我知道这个世人不会感开了我
你我在等待
不会怕我的感觉你说我在我的手 一定有人都没不错
不要问你不要要的人
我想你我不要再知事
我也能不会再想
但我有人爱你的我爱
你说我说一点走
让我们在半边 的照跳我没有
我说的感觉 我会想你  不要我的微笑就想多快乐
我不会再要我想不得
这么多小丑你要想走
我只是你的感觉

3、test01

模型监控

训练、测试过程全记录

1、训练过程


2018-10-13 21:25:28.646563:
step: 10/10000...  loss: 6.4094...  0.2286 sec/batch
step: 20/10000...  loss: 6.1287...  0.2296 sec/batch
step: 30/10000...  loss: 6.1107...  0.2211 sec/batch
step: 40/10000...  loss: 5.8126...  0.2467 sec/batch
step: 50/10000...  loss: 5.6969...  0.2366 sec/batch
step: 60/10000...  loss: 5.6081...  0.2406 sec/batch
step: 70/10000...  loss: 5.7305...  0.2411 sec/batch
step: 80/10000...  loss: 5.6465...  0.2441 sec/batch
step: 90/10000...  loss: 5.4519...  0.2381 sec/batch
step: 100/10000...  loss: 5.4479...  0.2271 sec/batch
step: 110/10000...  loss: 5.4051...  0.2236 sec/batch
step: 120/10000...  loss: 5.5111...  0.2226 sec/batch
step: 130/10000...  loss: 5.4023...  0.2311 sec/batch
step: 140/10000...  loss: 5.3445...  0.2266 sec/batch
step: 150/10000...  loss: 5.5066...  0.2326 sec/batch
step: 160/10000...  loss: 5.3925...  0.2376 sec/batch
step: 170/10000...  loss: 5.4850...  0.2326 sec/batch
step: 180/10000...  loss: 5.3654...  0.2206 sec/batch
step: 190/10000...  loss: 5.4041...  0.2421 sec/batch
step: 200/10000...  loss: 5.3814...  0.2181 sec/batch

……

step: 6800/10000...  loss: 3.0295...  0.2416 sec/batch
step: 6810/10000...  loss: 3.0675...  0.2747 sec/batch
step: 6820/10000...  loss: 3.0043...  0.2667 sec/batch
step: 6830/10000...  loss: 3.0258...  0.2276 sec/batch
step: 6840/10000...  loss: 2.8450...  0.2888 sec/batch
step: 6850/10000...  loss: 3.0091...  0.2797 sec/batch
step: 6860/10000...  loss: 3.1547...  0.2888 sec/batch
step: 6870/10000...  loss: 3.1725...  0.3108 sec/batch
step: 6880/10000...  loss: 2.8332...  0.2236 sec/batch
step: 6890/10000...  loss: 2.9765...  0.2186 sec/batch
step: 6900/10000...  loss: 3.0040...  0.2306 sec/batch
step: 6910/10000...  loss: 3.2340...  0.2136 sec/batch
step: 6920/10000...  loss: 2.8041...  0.2988 sec/batch
step: 6930/10000...  loss: 2.9514...  0.2216 sec/batch
step: 6940/10000...  loss: 3.1620...  0.2236 sec/batch
step: 6950/10000...  loss: 3.0633...  0.2136 sec/batch
step: 6960/10000...  loss: 2.8951...  0.2667 sec/batch
step: 6970/10000...  loss: 3.0201...  0.2206 sec/batch
step: 6980/10000...  loss: 3.0102...  0.2737 sec/batch
step: 6990/10000...  loss: 2.9885...  0.2527 sec/batch
step: 7000/10000...  loss: 3.1201...  0.2477 sec/batch

……

step: 8300/10000...  loss: 2.8474...  0.2296 sec/batch
step: 8310/10000...  loss: 2.8842...  0.2386 sec/batch
step: 8320/10000...  loss: 3.0446...  0.2246 sec/batch
step: 8330/10000...  loss: 3.0165...  0.2537 sec/batch
step: 8340/10000...  loss: 3.1366...  0.2266 sec/batch
step: 8350/10000...  loss: 2.9173...  0.3058 sec/batch
step: 8360/10000...  loss: 2.8468...  0.2356 sec/batch
step: 8370/10000...  loss: 3.0512...  0.2406 sec/batch
step: 8380/10000...  loss: 2.7532...  0.2286 sec/batch
step: 8390/10000...  loss: 3.0108...  0.2136 sec/batch
step: 8400/10000...  loss: 3.0818...  0.2787 sec/batch
step: 8410/10000...  loss: 2.9988...  0.2406 sec/batch
step: 8420/10000...  loss: 2.7640...  0.3449 sec/batch
step: 8430/10000...  loss: 3.0735...  0.2356 sec/batch
step: 8440/10000...  loss: 2.9183...  0.3610 sec/batch
step: 8450/10000...  loss: 2.9278...  0.3168 sec/batch
step: 8460/10000...  loss: 3.1321...  0.3660 sec/batch
step: 8470/10000...  loss: 2.9080...  0.2547 sec/batch
step: 8480/10000...  loss: 2.7861...  0.3108 sec/batch
step: 8490/10000...  loss: 3.0054...  0.2878 sec/batch
step: 8500/10000...  loss: 2.9389...  0.2366 sec/batch

……

step: 9800/10000...  loss: 2.8674...  0.3930 sec/batch
step: 9810/10000...  loss: 2.8695...  0.2136 sec/batch
step: 9820/10000...  loss: 2.8561...  0.2356 sec/batch
step: 9830/10000...  loss: 2.8824...  0.2186 sec/batch
step: 9840/10000...  loss: 2.8722...  0.2386 sec/batch
step: 9850/10000...  loss: 2.7315...  0.2116 sec/batch
step: 9860/10000...  loss: 2.8568...  0.2507 sec/batch
step: 9870/10000...  loss: 2.9153...  0.2216 sec/batch
step: 9880/10000...  loss: 2.9715...  0.2296 sec/batch
step: 9890/10000...  loss: 2.6941...  0.2256 sec/batch
step: 9900/10000...  loss: 2.7294...  0.3038 sec/batch
step: 9910/10000...  loss: 2.8968...  0.2376 sec/batch
step: 9920/10000...  loss: 2.9734...  0.2426 sec/batch
step: 9930/10000...  loss: 2.6401...  0.2276 sec/batch
step: 9940/10000...  loss: 2.8176...  0.2767 sec/batch
step: 9950/10000...  loss: 2.9532...  0.3429 sec/batch
step: 9960/10000...  loss: 2.9369...  0.2166 sec/batch
step: 9970/10000...  loss: 2.7590...  0.2206 sec/batch
step: 9980/10000...  loss: 2.8109...  0.2446 sec/batch
step: 9990/10000...  loss: 2.8011...  0.3038 sec/batch
step: 10000/10000...  loss: 2.8250...  0.3660 sec/batch

训练的数据集

1、训练的数据集为很多歌词

夜的第七章
打字机继续推向接近事实的那下一行
石楠烟斗的雾
飘向枯萎的树
沉默地对我哭诉
贝克街旁的圆形广场
盔甲骑士背上
鸢尾花的徽章 微亮
无人马车声响
深夜的拜访
邪恶在 维多利亚的月光下
血色的开场
消失的手枪
焦黑的手杖
融化的蜡像谁不在场
珠宝箱上 符号的假象
矛盾通往 他堆砌的死巷
证据被完美埋葬
那嘲弄苏格兰警场 的嘴角上扬
如果邪恶 是华丽残酷的乐章
它的终场 我会亲手写上
晨曦的光 风干最后一行忧伤
黑色的墨 染上安详
事实只能穿向
没有脚印的土壤
突兀的细微花香
刻意显眼的服装
每个人为不同的理由戴着面具说谎
动机也只有一种名字 那 叫做欲望
fafade~~fade~~fafa~
dedefa~~fade~~fafa~~
越过人性的沼泽
谁真的可以不被弄脏
我们可以 遗忘 原谅 但必须 知道真相
被移动过的铁床 那最后一块图终于拼上
我听见脚步声 预料的软皮鞋跟
他推开门 晚风晃了 煤油灯 一阵
打字机停在凶手的名称我转身
西敏寺的夜空 开始沸腾
在胸口绽放 艳丽的 死亡
我品尝着最后一口甜美的 真相
微笑回想正义只是安静的伸张
提琴在泰晤士
如果邪恶 是华丽残酷的乐章
它的终场 我会亲手写上
黑色的墨 染上安详
如果邪恶
是华丽残酷的乐章
它的终场 我会亲手写上
晨曦的光
风干最后一行忧伤
黑色的墨染上安详
黑色的墨染上安详
小朋友 你是否有很多问号
为什么 别人在那看漫画
我却在学画画 对着钢琴说话
别人在玩游戏 我却靠在墙壁背我的ABC
我说我要一台大大的飞机
但却得到一只旧旧录音机
为什么要听妈妈的话
长大后你就会开始懂了这段话 哼
长大后我开始明白
为什么我 跑得比别人快
飞得比别人高
将来大家看的都是我画的漫画
大家唱的都是 我写的歌
妈妈的辛苦 不让你看见
温暖的食谱在她心里面
有空就多多握握她的手
把手牵着一起梦游
听妈妈的话 别让她受伤
想快快长大 才能保护她
美丽的白发 幸福中发芽
天使的魔法 温暖中慈祥
在你的未来 音乐是你的王牌
拿王牌谈个恋爱
而我不想把你教坏
还是听妈妈的话吧
晚点再恋爱吧
我知道你未来的路
但妈比我更清楚
你会开始学其他同学
在书包写东写西
但我建议最好写妈妈
我会用功读书
用功读书 怎么会从我嘴巴说出
不想你输 所以要叫你用功读书
妈妈织给你的毛衣 你要好好的收着
因为母亲节到的时候我要告诉她我还留着
对了！我会遇到了周润发
所以你可以跟同学炫耀
赌神未来是你爸爸
我找不到 童年写的情书
你写完不要送人
因为过两天你会在操场上捡到
你会开始喜欢上流行歌
因为张学友开始准备唱《吻别》
听妈妈的话 别让她受伤
想快快长大 才能保护她
美丽的白发 幸福中发芽
天使的魔法 温暖中慈祥
听妈妈的话 别让她受伤
想快快长大 才能保护她
长大后我开始明白
为什么我 跑得比别人快
飞得比别人高
将来大家看的都是我画的漫画
大家唱的都是 我写的歌
妈妈的辛苦 不让你看见
温暖的食谱在她心里面
有空就多多握握她的手
把手牵着一起梦游
听妈妈的话 别让她受伤
想快快长大 才能保护她
美丽的白发 幸福中发芽
天使的魔法 温暖中慈祥

人工智能为你写歌词(写给陈奕迅)——基于TF利用RNN算法实现【机器为你作词】、训练&测试过程全记录

输出结果

1、test01

你的背包
一个人过我 谁不属了
不甘心 不能回头

我的背包载管这个
谁让我们是要不可
但求跟你过一生

你把我灌醉
即使嘴角从来未爱我
煽到你脖子
谁能凭我的比我
无赏
其实我的一切麻痹 我听过
不能够大到
爱人没有用
你想去哪里
如果美好不变可以
我会珍惜我最爱

我想将
鼓励爱你的
为何爱你不到你

我会加油工作争取享受和拼搏
三餐加一宿光档也许会寂寞
你想将
双手的温暖附托是你不知
但无守没抱过
不影响你不敢哭
其实没有火花 没抱动
不能够沉重
从来未休疚不够你不会
请你　这些眼发后

没有手机的日子

2、test02

谁来请你坐
全为你分声不需可怕
没有人机有几敷衍过的
难道再侣 被不想去为你
不如这样
你不爱你的
没有人歌颂
全一边扶暖之远一天一百万人
拥有殿军我想到我多 你真爱
从来未爱我　我们在
我有我是我们憎我 我不属于我
我想将

你的背包
原来不能回到你一起
难道我是谁也爱你 要不属力 不知道明年今日
明年今日 不是我不得到
爱你的背上我要
从来未不肯会 就要你不有

我会拖手会不会
难道我跟我眼睛的错　
全为这世上谁是你
不够爱你的汗
谁能来我的比你闷
不具名的演员没得到手的故事
不够含泪一个人 这么迂会
我们在

3、test03

谁来请你坐
全为你分声不需可怕
没有人机有几敷衍过的
难道再侣 被不想去为你
不如这样
你不爱你的
没有人歌颂
全一边扶暖之远一天一百万人
拥有殿军我想到我多 你真爱
从来未爱我　我们在
我有我是我们憎我 我不属于我
我想将

你的背包
原来不能回到你一起
难道我是谁也爱你 要不属力 不知道明年今日
明年今日 不是我不得到
爱你的背上我要
从来未不肯会 就要你不有

我会拖手会不会
难道我跟我眼睛的错　
全为这世上谁是你
不够爱你的汗
谁能来我的比你闷
不具名的演员没得到手的故事
不够含泪一个人 这么迂会
我们在

模型监控

训练、测试过程全记录

1、训练过程

2018-10-14 07:31:33.515130: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
step: 10/10000...  loss: 5.9560...  0.1500 sec/batch
step: 20/10000...  loss: 5.8495...  0.1300 sec/batch
step: 30/10000...  loss: 5.6970...  0.1600 sec/batch
step: 40/10000...  loss: 5.5893...  0.1300 sec/batch
step: 50/10000...  loss: 5.4582...  0.1300 sec/batch
step: 60/10000...  loss: 5.3003...  0.1400 sec/batch
step: 70/10000...  loss: 5.2871...  0.1600 sec/batch
step: 80/10000...  loss: 5.3841...  0.1500 sec/batch
step: 90/10000...  loss: 5.2470...  0.1400 sec/batch
step: 100/10000...  loss: 5.3061...  0.1500 sec/batch

……

step: 950/10000...  loss: 4.0972...  0.1600 sec/batch
step: 960/10000...  loss: 3.9542...  0.1900 sec/batch
step: 970/10000...  loss: 4.0406...  0.1500 sec/batch
step: 980/10000...  loss: 4.1385...  0.1400 sec/batch
step: 990/10000...  loss: 3.9897...  0.1600 sec/batch
step: 1000/10000...  loss: 3.9653...  0.1400 sec/batch
step: 1010/10000...  loss: 4.0501...  0.1300 sec/batch
step: 1020/10000...  loss: 3.9391...  0.1200 sec/batch
step: 1030/10000...  loss: 4.1195...  0.1400 sec/batch
step: 1040/10000...  loss: 3.9310...  0.1300 sec/batch
step: 1050/10000...  loss: 3.8972...  0.1200 sec/batch
step: 1060/10000...  loss: 3.9801...  0.1200 sec/batch
step: 1070/10000...  loss: 4.0620...  0.1200 sec/batch
step: 1080/10000...  loss: 3.8817...  0.1200 sec/batch
step: 1090/10000...  loss: 3.9839...  0.1301 sec/batch
step: 1100/10000...  loss: 3.9646...  0.1479 sec/batch

……

step: 4980/10000...  loss: 2.8199...  0.1200 sec/batch
step: 4990/10000...  loss: 2.9057...  0.1200 sec/batch
step: 5000/10000...  loss: 2.8073...  0.1300 sec/batch
step: 5010/10000...  loss: 2.6680...  0.1200 sec/batch
step: 5020/10000...  loss: 2.7442...  0.1200 sec/batch
step: 5030/10000...  loss: 2.7590...  0.1300 sec/batch
step: 5040/10000...  loss: 2.6470...  0.1300 sec/batch
step: 5050/10000...  loss: 2.7808...  0.1200 sec/batch
step: 5060/10000...  loss: 2.7322...  0.1200 sec/batch
step: 5070/10000...  loss: 2.8775...  0.1200 sec/batch
step: 5080/10000...  loss: 2.8139...  0.1200 sec/batch
step: 5090/10000...  loss: 2.7857...  0.1200 sec/batch
step: 5100/10000...  loss: 2.7652...  0.1200 sec/batch
step: 5110/10000...  loss: 2.8216...  0.1200 sec/batch
step: 5120/10000...  loss: 2.8843...  0.1200 sec/batch
step: 5130/10000...  loss: 3.0093...  0.1300 sec/batch
step: 5140/10000...  loss: 2.7560...  0.1200 sec/batch
step: 5150/10000...  loss: 2.7263...  0.1200 sec/batch
step: 5160/10000...  loss: 2.8014...  0.1200 sec/batch
step: 5170/10000...  loss: 2.7410...  0.1200 sec/batch
step: 5180/10000...  loss: 2.7335...  0.1200 sec/batch
step: 5190/10000...  loss: 2.8362...  0.1200 sec/batch
step: 5200/10000...  loss: 2.6725...  0.1300 sec/batch

……

step: 9690/10000...  loss: 2.3264...  0.1463 sec/batch
step: 9700/10000...  loss: 2.5150...  0.1425 sec/batch
step: 9710/10000...  loss: 2.3348...  0.1200 sec/batch
step: 9720/10000...  loss: 2.4240...  0.1277 sec/batch
step: 9730/10000...  loss: 2.4282...  0.1293 sec/batch
step: 9740/10000...  loss: 2.5858...  0.1232 sec/batch
step: 9750/10000...  loss: 2.2951...  0.1305 sec/batch
step: 9760/10000...  loss: 2.3257...  0.1263 sec/batch
step: 9770/10000...  loss: 2.4495...  0.1253 sec/batch
step: 9780/10000...  loss: 2.4302...  0.1289 sec/batch
step: 9790/10000...  loss: 2.5102...  0.1299 sec/batch
step: 9800/10000...  loss: 2.8486...  0.1254 sec/batch

……

step: 9900/10000...  loss: 2.4408...  0.1330 sec/batch
step: 9910/10000...  loss: 2.5797...  0.1275 sec/batch
step: 9920/10000...  loss: 2.4788...  0.1384 sec/batch
step: 9930/10000...  loss: 2.3162...  0.1312 sec/batch
step: 9940/10000...  loss: 2.3753...  0.1324 sec/batch
step: 9950/10000...  loss: 2.5156...  0.1584 sec/batch
step: 9960/10000...  loss: 2.4312...  0.1558 sec/batch
step: 9970/10000...  loss: 2.3816...  0.1279 sec/batch
step: 9980/10000...  loss: 2.3760...  0.1293 sec/batch
step: 9990/10000...  loss: 2.3829...  0.1315 sec/batch
step: 10000/10000...  loss: 2.3973...  0.1337 sec/batch

训练的数据集

1、训练的数据集为林夕写给陈奕迅的歌词，来源于网络

陈奕迅 - 梦想天空分外蓝
一天天的生活
一边怀念 一边体验
刚刚说了再见 又再见
一段段的故事
一边回顾 一边向前
别人的情节总有我的画面
只要有心就能看见
从白云看到 不变蓝天
从风雨寻回 梦的起点
海阔天空的颜色
就像梦想那么耀眼
用心就能看见
从陌生的脸 看到明天
从熟悉经典 翻出新篇
过眼的不只云烟
有梦就有蓝天
相信就能看见
美梦是个气球
签在手上 向往蓝天
不管高低不曾远离 我视线
生命是个舞台
不用排练 尽情表演
感动过的片段百看不厌
只要有心就能看见
从白云看到 不变蓝天
从风雨寻回 梦的起点
海阔天空的颜色
就像梦想那么耀眼
用心就能看见
从陌生的脸 看到明天
从熟悉经典 翻出新篇
过眼的不只云烟
相信梦想就能看见
有太多一面之缘 值得被留恋
总有感动的事 等待被发现
梦想天空分外蓝 今夕何年
Oh 看不厌
用心就能看见
从白云看到 不变蓝天
从风雨寻回 梦的起点
海阔天空的颜色
就像梦想那么耀眼
用心就能看见
从陌生的脸 看到明天
从熟悉经典 翻出新篇
过眼的不只云烟
有梦就有蓝天
相信就能看见

美梦是个气球
签在手上 向往蓝天
不管高低不曾远离 我视线
梦想是个诺言
记在心上 写在面前
因为相信 所以我看得见

人工智能为你写代码——基于TF利用RNN算法实现生成编程语言代码(C++语言)、训练&测试过程全记录

输出结果

1、test01

,&&curses = 0;
        if (tsk->state_perf_event_set && id_state_init == 1)
                return 0;
        if (!still &= is_stop_init(trace, context) {
                if (seq_context_trace_set_torture)
                        prev(size);
                pr_cont(&set);
                if (event->pid == 0)
                        goto out;

                return ret;
                return 0;
        }

        if (enabled == 0)
                ster_state = sched(ts, &sec_stack, task->tos);
}

static void sig_call(struct cpu_state *tr,
                                         struct rcu_state *size)
{
        int find_task;
        int cpu = 0;
        unsigned long free;

        set_timerective(task_parent_tasks);
        if (!perf_points == NULL)
                return;
        seq_printf(s, "  %ns\n", str, signal, flags);
        pid = ret;
}

static unsigned long table_read_id_chorch_set_calc(struct trace_policy *rsp)
{
        struct ftrace_event *init = task_sched_cpu();
        struct trace *ret = &ret;
        struct trace_iterator *trace_read_interval;

        sched = cpu_post_context(cpu);
        return -ENOMEM;
}

/**
 * time_context_tasks:  info as an set aling
 * @pron_set on this caller the at of this.
 *
 * Which ops set size that commitime is all can a is alloc

2、test02

.timespec */
        ring_buffer_init(resoulc->system_trace_context);
        size_of != cpu_cpu_stack;
        sprevert_return_init(&task_tail_timer, file);
        return ret;
}

static void statist_child_signal(struct task_struct *case,
                                 s instats_start *size,
                           struct ping *timer, char *stat)
{
        return init(&struct state + size, state->timeolt, str, cpu) = rq->to_count_compole_print_task_caller(cpu);
}

static int sched_copy_praid_irq(struct spin_lock *str, struct rq *rq->seq)
{
        if (task_const int)
                return;

        return;

static void rcu_boost_real(struct task_struct *stat, struct ftrace_event_call *trace)
{
        if (!page->temp_pid(&ts->class == 0) & TASK_MONTED; i < stop_pool(current, ftrace_console_file, trace_task_furcs)
                cred->flags & CANDER | 1

3、test03

.   start, arg state task
         * an the struct strings sigre the clear that and the secs. */

        return 0;
}

/*
 * Rincend time a current the arched on the for and the time
 * store the task tracer an a to in the set_tryset is interrupted
 * aling are a set the task trigger already ticks to call.  If call. The curr the check to string of
 * instatt is in a are the file set some try instructy it strentity of
 * is it as that the to son to inticate the state on the states at
         */
static int seq_buf_call(sched_lock)
{
        struct cpu *completion;

        /* The this stored the current allocation the tracer */
        if (timeore-- != 1)
                return 0;

        raw_buffer_state = true;
        context->trace_seq_puttour_spres(size,
                                                              seq->free_stres) {
                if (raw_int_risk(&regs) ||
                          struct class_cpu())
                        case TRACE_STACK_PONRINLED(tracing_time_clocks);
        }
        if (ret)
                count = set_test_size();

        trace_buffer_clear_restart(&call->filter);
        init_stop(timer, system, flags);
}

static int proc_system_trac

监控模型

训练&测试过程全记录

1、训练过程


2018-10-13 18:47:32.811423: 
step: 10/20000...  loss: 3.7245...  0.4903 sec/batch
step: 20/20000...  loss: 3.6450...  0.4442 sec/batch
step: 30/20000...  loss: 3.8607...  0.4482 sec/batch
step: 40/20000...  loss: 3.5700...  0.4993 sec/batch
step: 50/20000...  loss: 3.5019...  0.5444 sec/batch
step: 60/20000...  loss: 3.4688...  0.4512 sec/batch
step: 70/20000...  loss: 3.3954...  0.5244 sec/batch
step: 80/20000...  loss: 3.3532...  0.4031 sec/batch
step: 90/20000...  loss: 3.2842...  0.6517 sec/batch
step: 100/20000...  loss: 3.1934...  0.4893 sec/batch

……

step: 990/20000...  loss: 2.0868...  0.4111 sec/batch
step: 1000/20000...  loss: 2.0786...  0.4001 sec/batch
step: 1010/20000...  loss: 2.0844...  0.4352 sec/batch
step: 1020/20000...  loss: 2.1136...  0.4402 sec/batch
step: 1030/20000...  loss: 2.1023...  0.3199 sec/batch
step: 1040/20000...  loss: 2.0460...  0.4522 sec/batch
step: 1050/20000...  loss: 2.1545...  0.4432 sec/batch
step: 1060/20000...  loss: 2.1058...  0.3680 sec/batch
step: 1070/20000...  loss: 2.0850...  0.4201 sec/batch
step: 1080/20000...  loss: 2.0811...  0.4682 sec/batch
step: 1090/20000...  loss: 2.0438...  0.4301 sec/batch
step: 1100/20000...  loss: 2.0640...  0.4462 sec/batch
……
step: 1990/20000...  loss: 1.9070...  0.4442 sec/batch
step: 2000/20000...  loss: 1.8854...  0.3920 sec/batch

……

step: 4000/20000...  loss: 1.7408...  0.4612 sec/batch
step: 4010/20000...  loss: 1.8354...  0.4402 sec/batch
step: 4020/20000...  loss: 1.8101...  0.3951 sec/batch
step: 4030/20000...  loss: 1.8578...  0.4422 sec/batch
step: 4040/20000...  loss: 1.7468...  0.3770 sec/batch
step: 4050/20000...  loss: 1.8008...  0.4301 sec/batch
step: 4060/20000...  loss: 1.9093...  0.3650 sec/batch
step: 4070/20000...  loss: 1.8889...  0.4582 sec/batch
step: 4080/20000...  loss: 1.8673...  0.4682 sec/batch
step: 4090/20000...  loss: 1.7999...  0.3951 sec/batch
step: 4100/20000...  loss: 1.7484...  0.4582 sec/batch
step: 4110/20000...  loss: 1.7629...  0.4071 sec/batch
step: 4120/20000...  loss: 1.6727...  0.3940 sec/batch
step: 4130/20000...  loss: 1.7895...  0.3750 sec/batch
step: 4140/20000...  loss: 1.8002...  0.3860 sec/batch
step: 4150/20000...  loss: 1.7922...  0.4532 sec/batch
step: 4160/20000...  loss: 1.7259...  0.3951 sec/batch
step: 4170/20000...  loss: 1.7123...  0.4642 sec/batch
step: 4180/20000...  loss: 1.7262...  0.3760 sec/batch
step: 4190/20000...  loss: 1.8545...  0.3910 sec/batch
step: 4200/20000...  loss: 1.8221...  0.3539 sec/batch
step: 4210/20000...  loss: 1.8693...  0.4472 sec/batch
step: 4220/20000...  loss: 1.8502...  0.4793 sec/batch
step: 4230/20000...  loss: 1.7788...  0.3539 sec/batch
step: 4240/20000...  loss: 1.8240...  0.4793 sec/batch
step: 4250/20000...  loss: 1.7947...  0.4101 sec/batch
step: 4260/20000...  loss: 1.8094...  0.3630 sec/batch
step: 4270/20000...  loss: 1.7775...  0.4021 sec/batch
step: 4280/20000...  loss: 1.8868...  0.3950 sec/batch
step: 4290/20000...  loss: 1.7982...  0.4532 sec/batch
step: 4300/20000...  loss: 1.8579...  0.3359 sec/batch
step: 4310/20000...  loss: 1.7709...  0.4412 sec/batch
step: 4320/20000...  loss: 1.7422...  0.4011 sec/batch
step: 4330/20000...  loss: 1.7841...  0.5775 sec/batch
step: 4340/20000...  loss: 1.7253...  0.4532 sec/batch
step: 4350/20000...  loss: 1.8973...  0.3479 sec/batch
step: 4360/20000...  loss: 1.7462...  0.3680 sec/batch
step: 4370/20000...  loss: 1.8291...  0.5204 sec/batch
step: 4380/20000...  loss: 1.7276...  0.3930 sec/batch
step: 4390/20000...  loss: 1.7404...  0.3289 sec/batch
step: 4400/20000...  loss: 1.6993...  0.4462 sec/batch
step: 4410/20000...  loss: 1.8670...  0.3920 sec/batch
step: 4420/20000...  loss: 1.8217...  0.4301 sec/batch
step: 4430/20000...  loss: 1.8339...  0.5164 sec/batch
step: 4440/20000...  loss: 1.7154...  0.3660 sec/batch
step: 4450/20000...  loss: 1.8485...  0.3920 sec/batch
step: 4460/20000...  loss: 1.7758...  0.4161 sec/batch
step: 4470/20000...  loss: 1.7017...  0.5234 sec/batch
step: 4480/20000...  loss: 1.6939...  0.3379 sec/batch
step: 4490/20000...  loss: 1.7715...  0.3951 sec/batch
step: 4500/20000...  loss: 1.7940...  0.4492 sec/batch
step: 4510/20000...  loss: 1.7804...  0.3740 sec/batch
step: 4520/20000...  loss: 1.7876...  0.5073 sec/batch
step: 4530/20000...  loss: 1.7149...  0.5825 sec/batch
step: 4540/20000...  loss: 1.7723...  0.3961 sec/batch
step: 4550/20000...  loss: 1.8180...  0.4271 sec/batch
step: 4560/20000...  loss: 1.7757...  0.4933 sec/batch
step: 4570/20000...  loss: 1.8858...  0.3309 sec/batch
step: 4580/20000...  loss: 1.7332...  0.3890 sec/batch
step: 4590/20000...  loss: 1.8466...  0.4251 sec/batch
step: 4600/20000...  loss: 1.8532...  0.3930 sec/batch
step: 4610/20000...  loss: 1.8826...  0.3850 sec/batch
step: 4620/20000...  loss: 1.8447...  0.3359 sec/batch
step: 4630/20000...  loss: 1.7697...  0.4221 sec/batch
step: 4640/20000...  loss: 1.9220...  0.3549 sec/batch
step: 4650/20000...  loss: 1.7555...  0.4011 sec/batch
step: 4660/20000...  loss: 1.8541...  0.3830 sec/batch
step: 4670/20000...  loss: 1.8676...  0.4181 sec/batch
step: 4680/20000...  loss: 1.9653...  0.3600 sec/batch
step: 4690/20000...  loss: 1.8377...  0.3981 sec/batch
step: 4700/20000...  loss: 1.7620...  0.4291 sec/batch
step: 4710/20000...  loss: 1.7802...  0.4251 sec/batch
step: 4720/20000...  loss: 1.7495...  0.4131 sec/batch
step: 4730/20000...  loss: 1.7338...  0.3299 sec/batch
step: 4740/20000...  loss: 1.9160...  0.4662 sec/batch
step: 4750/20000...  loss: 1.8142...  0.3389 sec/batch
step: 4760/20000...  loss: 1.8162...  0.3680 sec/batch
step: 4770/20000...  loss: 1.8710...  0.4552 sec/batch
step: 4780/20000...  loss: 1.8923...  0.4321 sec/batch
step: 4790/20000...  loss: 1.8062...  0.4061 sec/batch
step: 4800/20000...  loss: 1.8175...  0.4342 sec/batch
step: 4810/20000...  loss: 1.9355...  0.3459 sec/batch
step: 4820/20000...  loss: 1.7608...  0.4191 sec/batch
step: 4830/20000...  loss: 1.8031...  0.3991 sec/batch
step: 4840/20000...  loss: 1.9261...  0.4472 sec/batch
step: 4850/20000...  loss: 1.7129...  0.3981 sec/batch
step: 4860/20000...  loss: 1.7748...  0.4642 sec/batch
step: 4870/20000...  loss: 1.8557...  0.4221 sec/batch
step: 4880/20000...  loss: 1.7181...  0.4452 sec/batch
step: 4890/20000...  loss: 1.7657...  0.5134 sec/batch
step: 4900/20000...  loss: 1.8971...  0.4813 sec/batch
step: 4910/20000...  loss: 1.7947...  0.3670 sec/batch
step: 4920/20000...  loss: 1.7647...  0.4362 sec/batch
step: 4930/20000...  loss: 1.7945...  0.3509 sec/batch
step: 4940/20000...  loss: 1.7773...  0.4342 sec/batch
step: 4950/20000...  loss: 1.7854...  0.4121 sec/batch
step: 4960/20000...  loss: 1.7883...  0.3720 sec/batch
step: 4970/20000...  loss: 1.7483...  0.3700 sec/batch
step: 4980/20000...  loss: 1.8686...  0.5645 sec/batch
step: 4990/20000...  loss: 1.8472...  0.2075 sec/batch
step: 5000/20000...  loss: 1.8808...  0.1955 sec/batch

……

step: 9990/20000...  loss: 1.7760...  0.2306 sec/batch
step: 10000/20000...  loss: 1.6906...  0.2256 sec/batch
……

step: 19800/20000...  loss: 1.5745...  0.2657 sec/batch
step: 19810/20000...  loss: 1.7075...  0.2326 sec/batch
step: 19820/20000...  loss: 1.5854...  0.3660 sec/batch
step: 19830/20000...  loss: 1.6520...  0.3529 sec/batch
step: 19840/20000...  loss: 1.6153...  0.3434 sec/batch
step: 19850/20000...  loss: 1.6174...  0.3063 sec/batch
step: 19860/20000...  loss: 1.6060...  0.2717 sec/batch
step: 19870/20000...  loss: 1.5775...  0.2627 sec/batch
step: 19880/20000...  loss: 1.6181...  0.2326 sec/batch
step: 19890/20000...  loss: 1.5117...  0.2547 sec/batch
step: 19900/20000...  loss: 1.5613...  0.2356 sec/batch
step: 19910/20000...  loss: 1.6465...  0.2346 sec/batch
step: 19920/20000...  loss: 1.5160...  0.2607 sec/batch
step: 19930/20000...  loss: 1.6922...  0.2306 sec/batch
step: 19940/20000...  loss: 1.8708...  0.2527 sec/batch
step: 19950/20000...  loss: 1.5579...  0.2276 sec/batch
step: 19960/20000...  loss: 1.5850...  0.2376 sec/batch
step: 19970/20000...  loss: 1.6798...  0.2286 sec/batch
step: 19980/20000...  loss: 1.5684...  0.2667 sec/batch
step: 19990/20000...  loss: 1.4981...  0.2617 sec/batch
step: 20000/20000...  loss: 1.5322...  0.3199 sec/batch

训练的数据集展示

1、数据集展示

/**
 * context_tracking_enter - Inform the context tracking that the CPU is going
 *                          enter user or guest space mode.
 *
 * This function must be called right before we switch from the kernel
 * to user or guest space, when it's guaranteed the remaining kernel
 * instructions to execute won't use any RCU read side critical section
 * because this function sets RCU in extended quiescent state.
 */
void context_tracking_enter(enum ctx_state state)
{
	unsigned long flags;

	/*
	 * Repeat the user_enter() check here because some archs may be calling
	 * this from asm and if no CPU needs context tracking, they shouldn't
	 * go further. Repeat the check here until they support the inline static
	 * key check.
	 */
	if (!context_tracking_is_enabled())
		return;

	/*
	 * Some contexts may involve an exception occuring in an irq,
	 * leading to that nesting:
	 * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit()
	 * This would mess up the dyntick_nesting count though. And rcu_irq_*()
	 * helpers are enough to protect RCU uses inside the exception. So
	 * just return immediately if we detect we are in an IRQ.
	 */
	if (in_interrupt())
		return;

	/* Kernel threads aren't supposed to go to userspace */
	WARN_ON_ONCE(!current->mm);

	local_irq_save(flags);
	if ( __this_cpu_read(context_tracking.state) != state) {
		if (__this_cpu_read(context_tracking.active)) {
			/*
			 * At this stage, only low level arch entry code remains and
			 * then we'll run in userspace. We can assume there won't be
			 * any RCU read-side critical section until the next call to
			 * user_exit() or rcu_irq_enter(). Let's remove RCU's dependency
			 * on the tick.
			 */
			if (state == CONTEXT_USER) {
				trace_user_enter(0);
				vtime_user_enter(current);
			}
			rcu_user_enter();
		}
		/*
		 * Even if context tracking is disabled on this CPU, because it's outside
		 * the full dynticks mask for example, we still have to keep track of the
		 * context transitions and states to prevent inconsistency on those of
		 * other CPUs.
		 * If a task triggers an exception in userspace, sleep on the exception
		 * handler and then migrate to another CPU, that new CPU must know where
		 * the exception returns by the time we call exception_exit().
		 * This information can only be provided by the previous CPU when it called
		 * exception_enter().
		 * OTOH we can spare the calls to vtime and RCU when context_tracking.active
		 * is false because we know that CPU is not tickless.
		 */
		__this_cpu_write(context_tracking.state, state);
	}
	local_irq_restore(flags);
}
NOKPROBE_SYMBOL(context_tracking_enter);
EXPORT_SYMBOL_GPL(context_tracking_enter);

void context_tracking_user_enter(void)
{
	context_tracking_enter(CONTEXT_USER);
}
NOKPROBE_SYMBOL(context_tracking_user_enter);

/**
 * context_tracking_exit - Inform the context tracking that the CPU is
 *                         exiting user or guest mode and entering the kernel.
 *
 * This function must be called after we entered the kernel from user or
 * guest space before any use of RCU read side critical section. This
 * potentially include any high level kernel code like syscalls, exceptions,
 * signal handling, etc...
 *
 * This call supports re-entrancy. This way it can be called from any exception
 * handler without needing to know if we came from userspace or not.
 */
void context_tracking_exit(enum ctx_state state)
{
	unsigned long flags;

	if (!context_tracking_is_enabled())
		return;

	if (in_interrupt())
		return;

	local_irq_save(flags);
	if (__this_cpu_read(context_tracking.state) == state) {
		if (__this_cpu_read(context_tracking.active)) {
			/*
			 * We are going to run code that may use RCU. Inform
			 * RCU core about that (ie: we may need the tick again).
			 */
			rcu_user_exit();
			if (state == CONTEXT_USER) {
				vtime_user_exit(current);
				trace_user_exit(0);
			}
		}
		__this_cpu_write(context_tracking.state, CONTEXT_KERNEL);
	}
	local_irq_restore(flags);
}
NOKPROBE_SYMBOL(context_tracking_exit);
EXPORT_SYMBOL_GPL(context_tracking_exit);

void context_tracking_user_exit(void)
{
	context_tracking_exit(CONTEXT_USER);
}
NOKPROBE_SYMBOL(context_tracking_user_exit);