Go与数据科学

我是丹尼尔·维特纳克，欢迎收听Go Time。

I'm Daniel Witenack, and this is Go Time.

Go Time是一档每周播出的播客，我们在这里讨论围绕Go编程语言、社区以及相关的一切有趣话题。

It's Go Time, a weekly podcast where we discuss interesting topics around the Go programming language, the community, and everything in between.

如果你目前正在使用Go，或者希望学习Go，那么这档节目非常适合你。

If you currently write Go or aspire to, this is the show for you.

欢迎再次回到Go Time的另一期节目。

Welcome back for another episode of Go Time.

这是第四期节目。

This is episode number four.

今天，我们将讨论Go在数据科学中的应用，以及我们本周发现的其他有趣新闻和项目。

Today, we'll be talking about Go for data science and other interesting news and projects we've come across this week.

我是埃里克·斯。

I'm Eric St.

马丁，和我一起的依然是我们另一位出色的主持人布莱恩·凯特尔森。

Martin, and with me as always are our other amazing host, Brian Kettleson.

打个招呼吧，Brian。

Say hello, Brian.

你好。

Hello.

还有Carlicia Campos。

And also, Carlicia Campos.

打个招呼吧，Carlicia。

Say hello, Carlicia.

很高兴来到这里。

Glad to be here.

嗨。

Hi.

对于今天已经收听的听众，我们今天还邀请到了Daniel Wightnack，他非常投入于Gopher数据科学领域。

And for anybody who's already listening today, we're joined by Daniel Wightnack, who is very vested in gopher data science.

他今年还将在GopherCon上就这一主题发表演讲。

He's also going to be speaking at GopherCon this year about it.

丹尼尔，你怎么样？

How are you, Daniel?

很好。

Good.

能来这里真是太好了。

Good good to be here.

那我们先来聊聊这周大家在新闻、文章或各种事情中遇到的有趣内容吧。

So let's kick this episode off by talking about anything we've kind of run across this week in news and just random articles and things.

大家都有什么发现？

What's everybody got?

这周最大的事情是 Go 1.7 的功能冻结。

Hey, the biggest thing this week is the feature freeze for Go 1.7.

这一消息已在 Golang Nuts 和 Golang Dev 邮件列表中公布。

It was announced on the Golang Nuts and Golang Dev mailing list.

这次更新涉及 SSA 和编译器的大量改动。

There's a lot of changes in this with the SSA and compiler changes.

所以，所有人在外面都需要下载 Go 1.7 并编译它，测试你们的程序，因为这是一个重大的变化，可能会导致程序出错。

So, everybody out there needs to download Go 1.7 and compile it, test your programs because this is a really big change, and there's opportunity for breakage.

当然，我们的测试覆盖了所有内容，但对我们来说这是一个很大的改动，编译速度也大幅下降了。

Of course, our tests cover everything, but it's a big place for us and the compile speeds are down, down, down, down.

我看到一些关于二进制文件的内容。

Saw something about binaries.

二进制文件中是否包含依赖包？

Is it vendor in binaries?

没来得及深入研究。

Didn't have a chance to investigate.

我非常感兴趣。

Was super interested.

有人知道吗？

Does anybody know?

我觉得是只有二进制的包，对吧？

Think it was binary only packages, right?

不提供源代码吗？

Without providing the source?

没错。

Exactly.

现在你可以获取一个仅包含二进制文件的包。

You can go you can go get a binary only package now.

所以你可以两种方式都做？

So you can do it both ways?

正确。

Correct.

如果你想发布一个仅包含二进制文件而没有源代码的预编译包，你可以通过提供一个指向 GoGet 的 URL 来实现，或者使用 GoGet。

If you wanted to to release a precompiled package that was only a binary with no source, you could do that by by giving somebody a URL to a GoGet or to use GoGet.

哦，我明白你的意思了。

Oh, I see what you're saying.

明白了。

Gotcha.

是的。

Yeah.

所以更像是像Windows平台上可以购买的商业软件包和库那样的东西。

So kind of more along the lines of, like, the commercial packages and libraries that you can buy, say, for the Windows side of the world.

那么，如果我使用你的包的二进制版本，我可以把它直接放进我的项目里并作为二进制文件使用吗？

So then if I'm using your bind your package as a binary, I can pop it in my project and use it as a binary?

我不确定这是否适用于预编译的库。

I don't know if it applies to precompiled libraries.

我觉得这只适用于命令，但看看它是否也适用于库文件（.a文件）会很有趣。

I I think it was only commands, but it would be interesting to see if it it actually applied to the, library files, the a files.

我得再深入研究一下。

I'll have to look more into that.

我只是偶然看到过一点关于这个的内容，当时匆匆扫了一眼。

I just I briefly saw something about that, and I kinda breezed over it.

这周我真的很忙。

I've been really busy this week.

但那个版本还有大约300个已关闭的工单。

But there's also there's something like 300 closed tickets for that release.

令人兴奋。

Exciting.

我期待编译速度的提升。

I'm excited for the compile times.

是啊，我也一样。

Yeah, me too.

我迫不及待想回到1.4的速度了。

I can't wait to get back to, 1.4 speeds.

我们越来越接近了。

We're getting closer.

虽然还是比1.4慢，但我们已经恢复了大部分丢失的性能，这让大家都很高兴。

It's still faster it's still slower than 1.4, but we've regained a lot of that lost territory, and that makes everybody happy.

这周我还看到了另一件有趣的事，就是布拉德和安德鲁的直播编码环节。

So the other interesting thing that I saw, this week too was, Brad and Andrew's live coding sessions.

你们有没有看到，他们正在招募想要参与实时代码审查的人？

Did you guys see that, where they're, putting a call out for people who want live code reviews?

这有点吓人。

Well, that's frightening.

我没这么说，但听起来很棒。

I didn't say that, but that sounds great.

是的。

Yeah.

我也没看到，但我非常想参与这个活动。

I didn't see it either, but I would love to I would love to participate in that.

我以前看过他们这么做，效果非常好。

I've seen it be I've seen them doing it in the past, and it's great.

我不确定。

I don't know.

如果你以前没有自我怀疑，想象一下布拉德和安德鲁把你的代码撕得粉碎。

If you didn't have impostor syndrome before, imagine Brad and Andrew tearing up your code.

你知道，他们俩都是很棒的人，别误会我的意思。

You know, they're both great guys, so don't get me wrong.

只是，你知道的，要让这群人满意可不容易。

It's just, you know, that is a tough crowd to please.

我不知道，但我觉得这只会让我对自己的看法暴露出来，对吧？

I don't know, think that it would just make my own thoughts of myself come to light, right?

就像，我知道，看吧？

Like, I knew it, see?

我知道自己是个糟糕的程序员。

I knew I was terrible coder.

Gotime.fm，我们花一个小时谈论自己的冒名顶替综合征。

Gotime.fm, where we talk about our imposter syndrome for an hour.

这周还有另一个大新闻，那就是彼得·贝尔贡的文章，关于他更新了2014年的Go最佳实践演讲。

There's another big, news item this week, and that was, Peter Bergon's article about, his update of his Go Best Practices talk from 2014.

他将内容更新到了2016年，里面有很多非常不错的收获。

He updated that to, 2016 and there's a lot of really good takeaways in there.

我觉得我们节目里没时间把它们都讲完。

I don't think we have time on the show to cover them all.

但里面确实有很多关于依赖管理、测试、如何组织你的库以及如何命名包的绝佳建议。

But, some really good tips in there about dependency management, about testing, how to structure your library, what to name your packages.

你想得到的，里面都整理得非常好。

You name it, it's really nicely in there.

就像我在推特上说的，我觉得这简直是经典。

Like I said on Twitter, think this is is really canon.

每个Go开发者都应该读一读这篇文章，并努力把这些规则内化，因为它们真的非常有价值。

This should be, every Go developer should read this and try to internalize all of these rules because they're really good stuff.

是的。

Yeah.

我觉得我要花点时间才能完全消化，但我觉得这是那种我应该收藏起来、每周都回头看看的东西。

Think it's gonna take a while for me to unpack, but I think it's one of those things, you know, I have bookmarked and I'm gonna visit every week.

当我实现不同功能时，我会参考它，看看自己是不是在以合理的方式做事，这会很有帮助。

And as I'm implementing different things, I think, you know, referencing that and just seeing if I'm doing things in a sane way is gonna be great.

是的。

Yeah.

我认为那场演讲至今仍然是我最喜爱的GopherCon 2014演讲。

Think that that talk was a favorite still to this day from, GopherCon twenty fourteen.

我真的很喜欢，看到他回去反思自己的想法，甚至否定一些他曾经认为正确的观点，这很有趣；不过我不确定他是否完全否定了2014年提出的任何概念。

Really So loved that it's it's fun to see him go back and kind of reflect on his thoughts and and to shoot down the things that he thought, or the way although I don't know whether he he fully dismissed any of the concepts that he presented in 2014.

我觉得他对一些观点变得没那么固执了，还新增了一些内容。

I think his his, he got a little looser on his beliefs on a couple of things, and then a couple of added new ones.

我需要再仔细读一遍。

I need to read through it again.

就像你说的，丹尼尔，你读一次会觉得：‘我刚读了什么？’

And like you said, Daniel, that it's like that you read it one time and you're like, what did I just read?

你得反复回去看。

You kind of have to keep going back to it.

是的。

Yeah.

这么大的内容需要很长时间才能完全内化。

Something this big takes a long time to internalize all of it.

我真希望能把这个演讲或者这篇文章，和一周前戴夫·钱尼关于错误处理的文章结合起来，每天晚上放在枕头上，希望我能一次性吸收所有这些内容。

I would love to take this talk or this this, this article, mix it with Dave Chaney's error handling article from a week or two ago, and, you know, just put it on my pillow and sleep on it every night, hoping that I could absorb all of that at once.

那太好了。

That'd be great.

我一直在克制自己，不想变成一个狂热的粉丝。

I'm holding myself back because I don't wanna be such a big fan girl.

但对于刚加入这个社区的人来说，彼得绝对是值得学习的对象。

But for people new to the community and to go, Peter is definitely somebody to absorb.

我认为彼得和戴夫·钱尼是我心目中最佳实践的前两名，是我认为非常值得追随的路径。

And I would say Peter and Dave Chaney are big on the top of my list as far as best practices and things that I think are very good paths to follow.

是的，我同意。

Yeah, I would agree.

而且我觉得，当我刚接触 Go 时，当然会去搜索各种 Stack Overflow 的回答之类的，但你会发现关于错误处理之类的问题，观点五花八门；而当我转向这些资源时，却为我在网络上看到的那些混乱观点带来了一些清晰的思路。

And I think coming into Go not too long ago, I think, of course, you search around different Stack Overflow responses and whatever and you can really get a mixed bag of ideas about, like Eric said, like about error handling or what have you, but kind of going to these resources I found gives some clarity on that mixed bag that you kind of find across the interweb.

我也想看看更多最佳实践之类的内容，以便做个比较和对照，对吧？

I'm interested to see some more best practices and stuff too, just to kind of compare and contrast, right?

因为我觉得人们对不同的方法有不同的成功经验，尤其是在依赖管理方面，这可能是个热门话题。

Because I think people have different success stories with different things, and especially when it comes to vendoring, that can be a hot topic.

我的意思是，我们完全可以花一整集来讨论依赖管理。

I mean, we could probably talk for a full episode about vendoring.

我觉得我们甚至可以办一场专门关于依赖管理的GopherCon，但仍然无法达成任何共识。

I think we could have an entire GopherCon on vendoring and still not get any consensus from anybody.

谁需要共识？

Who needs consensus?

没人需要共识。

Nobody needs consensus.

说得好。

Good point.

好吧。

All right.

那么继续，一些有趣的 Go 项目。

So moving on, interesting go projects.

说到睡觉和下载信息，布莱恩，你每晚都会下载一些酷炫的项目。

And speaking on sleeping and downloading information, Brian, with your nightly download of cool projects.

所以，几周前我看到了一个项目，当时我不太确定，因为我没看到全部的代码。

So here's here's one that I saw a couple weeks ago, and I wasn't too sure about it because I just I didn't I didn't see all of the code there.

但这个项目叫 Gafka，拼写是 g-a-f-k-a，是一套用于管理 Kafka 集群的 Go 工具。

But this one's called Gafka, g a f k a, and it's a suite of Go tools for managing Kafka clusters.

它在 github.com/funkygao/gafka，现在看起来非常完整了。

It's at github.com/funkygao/gafka, and it looks really complete now.

如果你管理 Kafka（很多人确实如此），这个项目非常有趣，它提供了一系列不错的 Go 命令和工具，帮你处理集群，避免了通常操作 Kafka 时不得不使用的繁琐的 Java 脚本。

Very interesting if you manage Kafka, which a lot of people do, and it's got a lot of nice little Go commands and Go tools for working with your cluster, things that will save you from, the ugly mounting of, Java shell scripts that you have to do generally when you're messing with Kafka.

所以，这些只是用于在 Kafka 世界中进行管理的辅助工具，还是替代了 Serama 之类的客户端库？

So now this is just helpers for administering in kind of that world for Kafka, or this replaces Serama and some of those libraries for client libraries?

我没看到里面有任何客户端库。

I didn't see any client libraries in there.

这只是一个用于管理集群、消费者组等的工具套件。

This is this is only a suite of tools for managing the cluster, at, consumer groups, that sort of thing.

所以里面至少有十几种不同的工具用于处理你的集群。

So lots of I I I want to say there are at least a dozen different utilities in there for dealing with your cluster.

现在，丹尼尔，我猜你在Kafka方面非常熟悉，毕竟你来自数据科学领域。

And now, Daniel, you're, I imagine you're quite versed in Kafka, being from the data science world.

是的。

Yeah.

我们在生产环境中使用Kafka，我认为，有了这个工具，我完全可以给那些负责Java脚本的家伙们点颜色看看。

We, we we use Kafka in production and, I think, you know, with this, I might be able to give those, Java Scholar guys that run for their money.

我还注意到它有一个功能，就是把健康信息上报到InfluxDB。

I also saw like, I I like the the part of it about, admitting health info to to InfluxDB.

这在我这边想象中会非常有用。

That's something I can can imagine super useful on my end.

好吧，还有别的吗？

All right, what else we got?

这周我从两个不同的地方接触到了聊天运维这个概念。

Well, I ran into this concept of chat ops this past week from two different places.

一个是远程 meetup 小组。

One was the remote meetup group.

他们请了一位嘉宾做了一场关于聊天运维的演讲，使用的是 Hubbots GitHub 项目。

They had somebody come in and do a talk about chat ops, and they did it using Hubbots GitHub hub GitHub.

Hubbot。

Hubbot.

我

I

我觉得应该读作 Hubbot。

think it's pronounced Hubbot.

我觉得发音就是这样。

Think that's how you pronounce it.

听起来没错。

That sounds right.

他做得非常出色。

And he did a marvelous job.

他解释了这是什么，进行了演示，并说明了如何使用它。

He said what it was, he demonstrated it, he did a demo, he explained how you can use it.

这太棒了。

It was fantastic.

我从未在使用过它的环境中工作过。

And I've never worked in a place that was using it.

我甚至不知道这是否非常新颖。

I don't even know if this is super new.

我知道聊天机器人的概念并不新，但将聊天运维直接融入你进行沟通的地方这个概念。

I know the concept of chatbots is not, but the concept of having chat ops right there where you are doing your communication.

然后我还遇到了这个微库，它是一个 Go 库。

And then I also came across this micro library, which is a Go library.

这是一个微服务工具包或库，具备聊天机器人功能。

It's a microservices kit or library, and it has capabilities for chatbots.

我很喜欢他们做了这个。

And I love it that they make it.

他们说他们有一个链接。

They say that they have a link.

他们有一篇博客文章，我们会提供链接。

They have a blog post, and we'll have a link to it.

他们说聊天机器人或聊天操作机器人应该成为第一等公民。

They said chatbots or chatops bots should be a first class citizen.

我觉得这很棒。

And I think it's great.

我经常在命令行上工作，我非常喜欢。

I work on the command line a lot and I love it.

但我也很喜欢将操作融入你与人交流的地方这个理念。

But I also love the concept of bringing operations into the place where you are communicating with people.

还有这些更新、更现代的工具所能提供的所有额外功能。

And also all the extra capabilities that you can get from these new, more modern tools.

你得观看远程会议的视频，才能明白我在说什么。

You have to watch the remote meetup video to see what I'm talking about.

我没法完全描述出来，但我用过这个工具，所以它就在我嘴边，就是说不出来。

I can't totally describe it, but I've worked with this, so it's not like, is at the tip of my tongue here to describe.

我只是感到非常着迷，我真的很想用这个工具来工作，让它成为现实。

I was just fascinated and I definitely would love to work using this tool so that it is.

是的。

Yeah.

我觉得吸引我的一部分原因，是逐渐远离了在 Slack 里不断复制粘贴终端截图到聊天频道的做法。

And I think part of what, what I see that that's appealing is, is kind of moving away from the being in Slack and, and, always, always copy pasting screenshots of my, of my terminal into, into the Slack channel.

而是通过聊天机器人直接执行命令，以一种人类易懂的方式将我的操作直接融入对话中，让所有人都在同一页面上，不再需要猜测我是在哪里执行的命令、我的环境是什么、为什么看起来这么奇怪，从而跨越这个障碍。

But rather, like, you know, executing the command via the chatbot and kind of having a human understandable version of what I'm doing and enter right into the conversation so everybody is on the same page and doesn't have to like wonder, you know, where where I executed that and what my environment is and why it looks weird and getting over that hurdle.

这很难，因为我曾经很喜欢聊天应用，但后来我越来越远离它们，因为我尽量不再像以前那样频繁查看聊天频道，因为实在太分散注意力了。

It's hard though because I used to love the chat app stuff and then I've been kind of pulled more and more away from it because I'm I try not to look in the chat channels nearly as often as I used to, because it gets really distracting.

这种干扰确实很烦人。

The distraction feels Yeah.

你看到通知后点进去一看，哦，原来是有人在部署东西。

You kinda see your notification and then you look in there and it's like, oh, it's just somebody, you know, deploying something.

没错，就是这样。

Yeah, exactly.

很高兴你提到这一点，因为我本来就想说，结果给忘了。

And I'm glad you brought that up because I wanted to mention that enough I forgot.

我在视频里看到的并不是通知，因为我们确实会收到通知，这些钩子也都存在。

What I saw on this video was not so it was not at all notifications because, yeah, we get notifications, we get these hookups and they are there.

这更多是关于额外的功能。

This was more about having extra features.

比如，查询某个内容并从你各个服务中获取响应。

For example, querying something and getting the response from different services across your services.

这更像是你处于主导地位，进行查询并请求信息，也许还有其他人也在场，你可以共享这些信息，而不是被动地接收信息推送。

It's more like you're in command and you're querying and you're asking for information and maybe somebody else is there as well and you can share the information as opposed to being like just a passive consumer, just like being dumped information.

我觉得这太酷了。

I thought that was extremely cool.

是的，我认为从框架的角度来看，Micro 是2016年Go语言世界中最令人感兴趣的东西，因为它提供了——这话说得有点大胆。

Yeah, I think from a framework perspective, micro is probably the most interesting thing in the Go world that's come out in 2016 because it offers- That's bold.

说得大胆，但你知道，我之前也说过Rails、Go和Docker。

Is bold, but you know, I called Rails, I called Go, I called Docker.

你得相信我这一次。

You're going to have to trust me on this one.

Micro 很重要。

Micro's big.

它之所以重要，是因为它提供了对微服务更广阔的视角。

And it's big because it offers a much broader view of microservices.

它不仅仅是一套帮助你构建微服务的工具，而是一个完整的生态系统。

It's not just a set of tools to help you make microservices, it's a whole ecosystem.

我认为这很重要，因为单独构建微服务并不如部署、管理和运营以及使用这些微服务来得重要。

And I think that's important because, making microservices on their own is not as important as deploying and managing and operating, and consuming those microservices.

因此，拥有这样一个庞大的框架和完整的工具生态系统，我认为会让微服务对每个人来说都更易用，也更易于管理。

So having that big framework, that whole ecosystem of tools, I think will make microservices more accessible to everyone and certainly more manageable.

聊天运维是它的一个很棒的方面，但这恰恰体现了Go接口的强大之处，因为你可以通过Protobuf或网页使用相同的接口与之交互。

The chat ops is a great side of it, but it just goes to show the power of Go's interfaces because it's the same interface that you can use to interact with it by protobuf or by, a web page.

因此，你可以使用微框架中的同一个微服务，通过命令行与之交互。

So you can take the same microservice in the micro framework and interact with it, from the command line.

你可以通过网页与之交互。

You can interact with it from a web page.

你可以通过Slack与之交互。

You can interact with it from Slack.

而这一切都使用了相同的接口。

And it's all using that same interface.

所以我认为这背后蕴含着巨大的潜力，你必须从GitHub开始关注它，因为这个项目前景广阔。

So I think there's a lot of power behind this, and, you need to start it in GitHub because this one's going places.

他正在招聘。

He's hiring.

今天早上我看到一条推文。

I I saw a tweet this morning.

阿西姆·阿斯兰是目前负责Micro项目的人，他正在招聘人才。

Asim Aslam is the guy who's, running Micro right now, and and he's looking for people.

太棒了。

Awesome.

我很期待看到它会发展成什么样。

I'm interested to see what comes out with it.

这周我遇到了一个叫Unique的项目，拼写挺特别的。

So one thing that I came across this week was a project called Unique, which is spelled interesting.

它是 u n I k。

It's u n I k.

我想这肯定不是我想到的唯一一种拼法。

I wouldn't be the first way I think to to spell it.

但我知道，至少布赖恩和我之前讨论过，容器化、微服务等技术背后的动机之一，就是想看看单内核（unikernels）什么时候能开始兴起。

But it's so one of the things I know that at least Brian and I have discussed before was kind of this whole motivation behind containerization and microservices and things of that nature was how long before kind of unikernels started taking off.

所以我发现了一个叫Unique的项目，它可以将你的Go应用编译成一个微内核，同时也支持Java、C和C++。

So one thing I came across was this project called Unique, which, allows you to compile your Go app, and it does Java and c and c plus plus as well, but compile it into a microkernel.

我还没试过，或者抱歉，是单内核。

And I've not yet played with it, but or I'm sorry, a unikernel.

所以你现在让我想到了微服务。

So you got micro on my mind here.

抱歉。

Sorry.

这是我的错。

It's my fault.

所以我很好奇，那里会出现什么样的成果，人们会怎么使用它，以及单内核会如何发展。

So I'm interested to see, what kinda comes along there and what people do with it and, how unikernels evolve.

但是

But

你们之前有使用过单内核的经验吗？

Do do you guys have previous experience with unikernels in general?

只是为了玩玩，构建过一些单内核，但除此之外没有了。

Only in in building some to play with for toy reasons, but I've Uh-huh.

是的。

Yeah.

至于生产环境的使用，我几乎没有过。

Not so much as far as, like, production usage.

只是

Just

我完全没接触过。

I don't at all.

所以，基本上，单内核的概念就是它是一个经过高度精简的内核，是一个可启动的程序。

So, yeah, basically, the notion of the unikernel is just it's a really trimmed down kernel that it's a bootable program.

对吧？

Right?

所以，当你看一个容器时，对吧，PID 1 启动的那个进程就是你的应用程序。

So so when you look at your container, right, like PID one, the process that starts up is your application.

但你仍然有一个操作系统来引导这个特定的进程。

So but you still have an operating system that bootstraps that particular process.

而单内核基本上就是你的应用程序就是操作系统。

And then a unikernel is basically your app is the operating system.

所以我认为它们在底层可能有一些关联的东西，但我也不是单内核专家，所以别太当真。

So and I think they have some stuff tied in under the covers, but I'm no unikernel expert either, so I wouldn't take my my word for it.

是的。

Yeah.

这是一个有趣的概念，我一直关注着想试试看，但还没真正去实践。

It's an interesting concept and something that's been on my radar to, to play around with, but, I haven't quite got there yet, so.

所以对我来说，有趣的地方在于，从安全角度来看，最安全的代码就是根本不存在的代码。

So I think the interesting thing for me is, you know, from the security world, the most secure code is the code that isn't there.

而单内核的理念就是，把你应用程序和与硬件通信所需的最小组件提取出来，编译成一个看起来像操作系统的东西来运行你的应用。

And the idea of a unikernel is that you take your app and only the pieces that you need to talk to the hardware and compile that down into something that that feels like an operating system to run your app.

因此，你的应用程序的攻击面大大缩小了，因为你没有像Ubuntu发行版那样包含所有多余的组件。

So the attack surface of your application is significantly smaller because you don't have all of the extraneous stuff that comes in an Ubuntu distribution, for example.

所以你只有一个端口是开放的，因为你只监听80端口。

So you've only got the one port open because you're only listening on port 80.

攻击面显著减小，我认为这是无内核的主要优势之一。

There's a significantly smaller attack surface, and I think that's one of the main benefits of unikernels.

至少在我看来是这样。

At least in my mind.

所以我认为我们这期节目还剩大约三十分钟。

So I think we have roughly thirty minutes left in the show.

那我们来和丹尼尔聊聊数据科学吧。

So let's, let's get chatting with Daniel about data science.

哦，这太令人兴奋了。

Oh, this is exciting.

我们已经好几个星期迫不及待想聊数据科学了。

We've been twitching for weeks to talk about data science.

这很重要。

This is big.

太棒了。

Sweet.

我很兴奋。

I'm excited.

布赖恩和我负责分析方面，不过不错。

Brian and I are on the analytics side of things, but Cool.

我的意思是，我们在这个领域已经花了四五年时间，专注于分析工作并开发支持他们的软件。

I mean, we've we spent a good, what, four, five years in this space working with analytics side and building software to support them.

所以处于边缘地带。

So on the fringe.

你不能称我们为数据科学家。

You you couldn't call us data scientists.

你可以称我们为赋能数据科学家的程序员。

You could call us the programmers that enable data scientists.

我们就是实习生。

We're the we're the interns.

没错。

That's right.

也许你可以为我们所有人先做个介绍。

Maybe you could start for all of us.

你知道吗？你能给我们简单介绍一下背景吗？

You know, can you give us kind of a background of of a primer?

什么是数据科学？

What is data science?

它处于什么位置？

Where does it fit?

哪些属于数据科学，哪些不属于数据科学？

What are the things that are data science and what are the things that aren't data science?

是的，我认为这确实是个很好的起点，因为根据你问的人不同，你会得到非常不同的答案。

Yeah, I think that's definitely a good place to start because there are depending on who you ask, you'll get very different answers.

我认为一个有用的区分标准是：当人们在推特上谈论数据科学时，他们指的是什么，而那些受雇为数据科学家的人实际上在行业中做些什么。

I think one kind of useful differentiator that I like to keep in mind is kind of what is data science and what are people talking about when they say data science on Twitter and then what are data scientists, those employed as data scientists actually doing in industry.

所以，我认为在第一种情况下，人们经常谈论的是像AlphaGo击败围棋大师这样的事情。

So I think in the first case, what people are talking about a lot are things like AlphaGo beating the Go grandmaster.

如果你不熟悉的话，AlphaGo 是由谷歌训练的一个深度神经网络，他们基本上训练这个神经网络，使其能够击败围棋世界冠军——这里的围棋是指棋盘游戏，而不是编程语言。

So if you're not familiar, AlphaGo is deep neural network trained by Google and they basically train this neural network such that it could beat the grandmaster Go champion, Go being the board game, not the programming language.

我觉得这真的非常了不起。

And I think that's this is really amazing.

这是一个巨大的成就，值得被广泛讨论。

This is a huge achievement and should be talked about a lot.

但另一方面，当你去看那些在业界受雇的数据科学家们，他们实际上在做什么呢？

But then on the other side of things, then you look at what do data scientists, most of them employed in industry, what are they doing?

最近《福布斯》上有一篇有趣的文章，对众多数据科学家进行了调查，了解了他们平时都在做些什么。

There was an interesting article recently in Forbes actually polled a bunch of data scientists and found out what they spend their time doing.

实际上，其中很大一部分——高达90%——都是在收集数据集，也就是获取数据、解析数据并从数据中提取模式。

So actually a lot of it, 90% of it, was gathering data sets, so collecting data sets, you know, parsing data and extracting patterns from data.

因此，数据科学家所做的工作约有90%并不一定涉及什么高深的机器学习，而更多是数据的搬运、转换和提取模式，使其变得有用。

So about 90% of what data scientists do doesn't necessarily involve some nifty machine learning, but more of the process of moving data around, transforming it, and extracting patterns to make it useful.

而剩下的那大约10%，则可能涉及各种其他事情。

Then that other kind of last 10% could be various things.

所以其中一小部分是做一些有趣的机器学习技术，另一小部分是制作仪表板或可视化，还有其他各种任务。

So a little sliver of that is doing kind of interesting machine learning techniques, maybe another sliver of that is making dashboards or visualizations and then various other tasks.

因此，总的来说，对我来说，数据科学就是将任何地方的数据转化为业务洞察的过程。

So I would say in general for me, data science is the process of transforming data wherever that might be into insights for your business.

有时候这可能需要神经网络，但几乎总是需要一些问题解决、一些ETL，也许还需要一些算术。

And sometimes that might require, you know, a neural network but pretty much always it requires some problem solving and some ETL and maybe some arithmetic.

所以我认为这些是数据科学的许多核心支柱。

So I think those are a lot of the, you know, kind of the pillars of data science.

我有个问题想问你，这个问题可能有点粗俗，提前抱歉。

So I've got a question for you and this is a little bit crass, so forgive me in advance.

但几年前，大概是两年前，我读到一篇文章，说塔吉特百货能比你更早知道你女儿怀孕了。

But I I read an article a couple of years ago, maybe two years ago, about how Target knows when your daughter's pregnant before you do.

没错。

Yep.

这这这

And it's it's it's

我妻子他们曾在Target工作，这确实是真的。

just My wife they worked for Target, it's apparently true.

是的。

Yeah.

这真把我吓坏了。

It it scared the hell out of me.

真的吓到我了。

It really did.

因为，你知道的，人们购买的都是每天都会买的东西，我从未想过他们能通过分析我购买的商品模式，与其他模式进行对比，从而发现：嘿，这人要离婚了。

Because, you know, people buying just the things that you buy every day, it it never occurred to me that they could take the the patterns of things that I buy, compare them to other patterns, and realize, hey, this guy's getting a divorce.

这个人怀孕了。

This one's pregnant.

这些人下个月就要生孩子了。

These guys are having a baby next month.

他们能通过这些数据了解我的生活。

They can tell by that data what my life is.

这几乎就像是手机世界中的元数据问题。

It's almost the metadata thing from the cell phone world.

这真的令人害怕。

It's really scary.

数据科学中有多少是受商业驱动的，又有多少是被广告推动的？

How much of data science is powered by commerce or is is, how much of data science is being pushed because of commerce and how much of it is being pushed because of advertising?

那么，这在整体格局中又处于什么位置呢？

Then, you know, where does that fit in the grand scheme of things?

推动数据科学进步的真正动力是什么？

What's the real driver behind the advances in data science?

是的，当然。

Yeah, sure.

所以我认为，如果我们回溯一下过去几年，某些行业其实一直以某种方式参与着数据科学。

So I think, you know, if we look a little bit back, you know, over the years, maybe, I mean, there's been certain industries that have always kind of been involved in data science in some sense.

如果我们看看保险公司或其他类似公司，它们关注的是根据个人经历等信息来评估风险。

If we look at insurance companies or other companies like that that are interested in assessing risk of a person based on things that have happened to them and other things like that.

这种情况已经持续了一段时间，后来人们逐渐意识到，我们可以针对特定类型的人群进行广告投放，因此在那段时间里，许多数据科学职位都与广告相关。

So that's been going on for a while and then there was kind of this time where people learned, oh we can certain types of people with advertising and so for a while there, you know, a lot of the data science job postings were advertising related.

现在我认为可以说，数据科学——或者说这种通过数据产生洞察的过程——已经渗透到企业各个层面。

Now I think it would be fair to say that data science or, you know, again I'm calling data science this process of creating like data driven insights is really permeating pretty much every level of businesses.

因此，如今许多数据科学家的工作是为内部业务人员提供支持，而不仅仅是为了直接为公司创造更多利润，他们可能关注的是智能监控基础设施的故障，对吧？

So a lot of data scientists now are working to inform even internal business people, not even necessarily to make a company more money directly, they might be concerned about intelligently monitoring outages in their infrastructure, right?

然后，从这种最基础的层面，一直延伸到直接与盈利相关的流程，比如优化营销广告等各个方面。

And then it goes all the way to, from that kind of lowest level all the way to the directly applicable processes to the money side of things having to do with marketing improving your ads and all of that.

所以我认为，从最底层的后端流程，到面向外部世界的市场营销，你都能看到数据科学无处不在。

So I think from the very lowest level on the back end processes all the way up to marketing yourselves to the outside world, I think you're seeing data science permeate all of that.

因此，我认为现在的理念是，数据科学家加入公司时，目的并不一定是说‘我要来优化广告’或‘我要来预测风险或欺诈’之类，而是许多公司正在组建数据科学团队，并将这些团队嵌入到组织的各个部门，思考如何让组织每个层级的流程都实现数据驱动。

So I think the idea is really that data scientists now, you're employed and you're coming into a company, the idea is not necessarily to say, I'm going to come in and I'm going to optimize ads or I'm going to come in and I'm going to predict risk or predict fraud or whatever it is, but a lot of companies are building data science teams and embedding even embedding data science teams across the organization to say how do we make our processes at each level of our organization data driven.

你提到运维（DevOps）方面，这很有趣。

It's interesting that you mentioned the DevOps side of it.

我虽然没看完整场演讲，但几周前参加的一个会议上的幻灯片我看过。

I didn't know see the whole talk but I saw the slides on a conference that happened a week or two ago.

我记不清是哪一个了。

It escapes me which.

我想说这家公司是像Uber、Halo这样的大公司之一，他们谈论了如何利用数据科学来减少监控系统中的误报。

And I want to say the company was one of the bigger companies like Uber or Halo or something like that and they were talking about how they use data science to reduce the false positives in their monitoring systems.

他们根据监控的周期和模式，知道可以将90%的误报减少，从而在非真实故障期间将值班呼叫减少了90%。

They knew because of the cycles and the patterns of their monitoring, they could reduce the false positives, the pages 90% of the time, and that cut down their, their duty calls while they were in the middle of an outage that wasn't really an outage by 90%.

你知道，这背后有着实实在在的成本节约和好处。

You know, there's real cost savings, real benefits behind that.

是的，没错。

Yeah, exactly.

在这种情况下，你实际上是在优化你的业务流程和工程流程。

So in that case, you're really optimizing a lot around your business processes and your engineering processes.

是的，这实际上来自Uber。

Yeah, that's actually so that's from Uber.

这是Uber的系统Argos，也就是他们的告警系统。

That's Uber, their system Argos, which is their alerting system.

实际上，在其后端，他们用 Go 语言自研了时序数据库。

And actually on the back end of that, the time series database, they wrote internally in Go.

我当时参加了那个演讲，真的非常令人印象深刻。

It's really I was at that talk and it was super impressive.

我被他们在这项学习中投入的全部智能所震撼，真正实现了大幅降低误报率的显著成果。

I was kind of blown away by the, all the intelligence that they're putting into that learning and really making some pretty astounding gains in, in reducing their false positives.

哦，不错。

Oh nice.

那个系统有没有开源的部分？

Was that, was there any open source component to that?

他们已经发布过其中任何内容了吗？

Have they released any of that yet?

据我所知，还没有。

As far as I know, not.

我的意思是，希望他们未来会这么做，毕竟我们最近看到的趋势就是这样。

I mean hopefully they will kind of given the trends we've been seeing lately.

我的意思是，希望在某个时候他们能发布数据库、前端或者任何其他部分。

I mean, hopefully at some point they're able to release maybe the database or the front end or whatever.

目前，据我所知，至少我上次看到的时候，大部分内容仍然是内部的，但他们确实深入讨论了他们建模的逻辑，以及如何希望随着时间推移不断改进系统。

Right now, I think at least the last I saw, most of it was still internal, but they did discuss, give a pretty good discussion of kind of their logic around how they're modeling things and how they're hoping to kind of keep improving the system over time.

所以这确实是一场值得一看的演讲。

So it's definitely an interesting talk to see.

我不确定视频是否已经上传到网上，但我知道幻灯片是有的，

I'm not sure if the video is up online, know the slides are,

好的，我们一定会把这些内容放到节目笔记里。

so All right, we'll put those in the show notes for sure.

让我们花点时间谈谈行业常用工具，比如当前的语言和框架，你在Go这边看到哪些工具可能会被取代？

Let's talk for a moment about the kind of tools of the trade, kind of like what what what are the current tools of the trade, languages and frameworks, and what do you see those being replaced with on the Go side?

因为我知道，我们之前私下聊过这个问题，也对这些技术的演进感到兴奋。

Because I know that's something that, you know, we've chatted kind of back channel about that a bit, and our excitement on watching some of these things kind of evolve.

是的，当然。

Yeah, yeah, sure.

所以我认为，就在一年前，如果我去参加某个数据科学活动，一个常见的问题是：我该学 R 还是学 Python？

So I would say even just like a year ago, if I was to go to some data science event, one of the big questions was, you know, should I learn R or should I learn Python?

这些长期以来一直是这个领域的主要语言。

So these are like the big or traditionally, you know, have been the big players in this space.

因此，有很多优秀的工具。

So there's a lot of great tools.

我对 R 方面不太熟悉，但我一直在使用 Python。

I'm not as familiar on the R side, but I worked a lot in Python.

在 Python 方面，你有一整套数值计算和数据科学工具，比如 pandas、SciPy、NumPy 等等。

So, you know, on the Python side you've got a lot this whole suite of numerical and data science type tools like Python pandas and SciPy, NumPy, all of these things.

这种情况持续了一段时间。

And that was kind of how things went for a while.

然后，就在最近，我在同一个会议上听到了关于 Uber 报警系统的讨论，我明显感觉到社区中的态度发生了变化：人们开始问，我该学什么语言？

Then, I you know, recently, when I at that same conference where they discussed the Uber alerting stuff, I was seeing definitely much more of an attitude in the community about, well, you know, the question was posed, what language should I learn?

我原本以为他们会回答 R 或 Python，但发言者——恰好是 Slack 的 Josh Wills——却说，你应该都学，因为我每周都会用到 Python、JavaScript，他还提到了 Go。

And I think they were expecting to hear, you know, R or Python, but the answer that the speaker gave, who happened to be Josh Wills from Slack, he basically said, you know, learn them all because every week I use Python, I use JavaScript, I use And he mentioned Go as well.

我认为，而且我希望社区能够逐渐认识到，每种语言都有其独特的功能和适用场景。

And I think and I hope that the community is kind of opening up to see that each of these languages has unique capabilities and unique use cases.

我希望我们不要把这些语言都混为一谈，认为它们都在做同样的事情，而是要在它们特别有用的地方加以利用，并发挥各自的独特优势。

And the hope is that we don't kind of morph all of these languages together as all kind of doing the same thing, but we utilize them where particularly useful and utilize their unique features.

当然，还有一整套围绕所谓大数据技术的工具，比如 Apache Spark 和 Hadoop，这些大多是 Java 和 Scala 的应用程序。

And then of course there's a whole suite of tools around kind of big data technologies, quote unquote, like Apache Spark and Hadoop and these are mostly kind of Java Scala applications.

但即便如此，大约一年前，这方面主要还是 Java 和 Scala，而现在你看到的范围更广了，比如 InfluxDB、Uber 开发的时间序列数据库，甚至像 Pachyderm 这样有趣的项目，它们都是不基于 Java 和 Scala 的大数据框架，而是采用了 Go 和 Docker 等技术。

But then, even then, like, you know, a year ago or something, there was a lot on that side of things, it was Java Scala and now you're seeing a kind of broader range and there's things like, like, you know, looking at InfluxDB or the time series database from Uber that Uber worked on or even things like, Pacaderm is an interesting project that are big data frameworks that are not Java Scala based and utilizing technologies like Go and Docker.

所以我认为，目前社区正在发生一场转变。

So I think there is a shift going on right now in the community.

我知道，如果你们也注意到了这些变化的话。

I know if you guys have seen that, any of that as well.

我认为 Pachyderm 是我最近看到的最有趣的技术之一，因为它几乎将‘一切都可以在笔记本电脑上完成’的理念，从 Hadoop 带入了 Docker 世界。

I think Pachyderm is probably one of the more interesting things I've seen in a while because it almost takes that, you can do everything on your laptop, philosophy with Aach and Sed moves it into the Docker world.

所以，你不再是将 Unix 命令管道连接起来，而是将 Docker 容器彼此连接起来。

So instead of, piping Unix commands together, you're just piping Docker containers together.

这使得大数据工作流变得非常有趣，尤其是当你引入 Mesos 或 Kubernetes 这样的系统时，可以实现大规模的大数据编排。

And that's that makes an interesting big data workflow, especially when you add something like, Mesos or Kubernetes into the the mix where you've got, you know, large orchestrations of big data things happening.

听起来真的很有趣。

It sounds really interesting.

是的。

Yeah.

我的意思是，这些年来我见过一些项目，人们尝试用 Go 语言做文本解析和一些概率计算。

I mean, I've seen a few projects over the years where people have started trying to do, you know, text parsing and some probabilistic stuff in Go.

但当时我们为这个节目做研究时，遇到的那个项目叫什么来着？

But, I mean, what was that project we ran across when we were we were kind of researching this show?

那一页内容之庞大真是令人震惊。

And it was it was just astonishing how big that page was of

哦，对。

Oh, yeah.

用于数据科学。

For data science.