仅仅在一个epoch之后，模型似乎就已经学会了英语单词。虽然词组本身不具有意义，其中的词语毫无疑问已经是英文，只不过有一两个排列的错误。在第76个epoch之后，我得到的验证复杂度为3.076，样本看起来变成了这样：

　　president economic spokesman executive for securities was support to put used the sharelike the acquired who pla

　　再次的，毫无疑问这是英文，但是没多大意义。但是，验证复杂度跟第一个epoch相比，并没有多大的提升。

　　川普语录生成器

　　我在美国写下这篇文章，此时正有一场总统大选在进行。这一次选举的候选人有些不寻常，我们想看看字母级别的RNN能从他们的言论中学到什么。

　　我搜集了川普几次演讲的底稿。你可以在Github上获取（https://github.com/deeplearningathome/rnn_text_writer）。所有的这些演讲知识产权都属于川普，开奖，在这里仅用于语言建模研究使用。我一共使用了7篇讲稿，作为训练集，还有一个作为验证集。需要注意的是，这是一个非常非常小的训练数据集。训练集中包含的的单词比PTB数据库要小一个量级。如果模型能在如此小的数据库中学习一些东西，那简直太令人兴奋了。自然地，我会训练基于字母的而不是基于单词的模型。我还降低了所有字母，忽视掉除了字母之外的一些东西，以及标点符号。

　　以下是我使用的训练配置：

　　用于抽样的种子词组，我选择了“make america”。当然，随机的模型产生的是随机的字母集。但是，随后，我们开始得到一些东西：

　　a border will be a folk the everyn country, last an lobfer as no

　　在第四个Epoch之后（训练复杂度为5.486）。这不仅仅看起来更像英语，同时也反映了川普讲话的一些重点话题（比如，border, country）。再次的，在训练之前，模型对英语或者川普一无所知，仅仅有作为输入特征的字母。接近训练结束时（第55epoch，训练复杂度3.029），我得到如下的一些东西：

　　n plan will be again.the radical islamic steel that establishme

　　使用最后的模型，从中抽取更多的文本。你可以给予任何你想要的种子作为输入，抽取任意多的字母。

　　如果这样训练模型：

　　你可以通过添加抽样标志从中进行抽样：

　　那么，我们开始吧：

种子: “hillary”.

　　继续:

　　n will end speech taxes. theyre j2开奖直播istration almost choa. it is time to federal people.

种子: “russia”.

　　继续:

　　n, will not even work at our victory of as now? radical immigrants who start only has been me as federal erol trapped into being shongth mentions.

种子: “war on terror”.

　　继续:

　　ism around who are many earels bill closs institute and tonight, they need to couse today, she has been every, and secure the demainst tould came they end by the failed radical islam immigration

种子: “build a wall around mexico”.

　　继续:

　　is highated to have the transpacifical great illegal immigrants will be able her signing too j2开奖直播istration.this includes now is not a children, theyve reform everyant at our workers

种子: “our economy is”.

　　继续:

　　about pulpneedressed delieve nothing to already families under the american people who want americans. we will cause the shifted by the national illegal immigrants shes cannot fwart to protect our

最后，从RNN Donald更大的样本中，种子：“if i win election”

　　继续：

　　the hillary clinton cheated, and we have on our grussics and close.just that how for the convioration will be her its redicentable that. this includes immigration class nearly raxs activity, and terrorism will, sobificial politicians man at sofice. our new good. never runes to obamaclinton, just one who have provide the reprofuction, needs of corations, and brutally friend to do the american people, for it away. which theng me in it, the u.s.you cant not believe. we have an igned to speech. together, we are going to take a members, or doctors with system in order to every other agreements has been so served isis about how our provided. there is how surrend.remember, all up for the radical people their eeerly stuel million projects and crime.this are going to know about her alartpess of anysoming now talking hillary clintonwages?i have come and better east, and me tunn this taxard obama grow and the must make america many、

　　显然，你实际的样本会根据抽样过程固有的随机性而产生变化。

　　结论