Name: 什麼才是真正的數據科學？由數據科學家講述 (What REALLY is Data Science? Told by a Data Scientist)
Uploaded: 2021-01-14T08:11:51.000Z
Duration: 11 min 9 s
Description: 【看影片學英語】數萬部 YouTube 影片，搭配英漢字典即點即查，輕鬆掌握單字發音與用法，長久累積看電影不必再看字幕。

Data science is not about making complicated models. It's not about making awesome visualizations

進階形容詞比較級句型

It's not about writing code data science is about using data to create as much impact as possible for your company

Now impact can be in the form of multiple things

It could be in the form of insights in the form of data products or in the form of product recommendations for a company

Now to do those things, then you need tools like making complicated models or data visualizations or writing code

your job is to solve real company problems using data and what kind of tools you use we don't care

Now there's a lot of misconception about data science, especially on YouTube

and I think the reason for this is because there's a huge misalignment between

what's popular to talk about and what's needed in the industry. So because of that I want to make things clear. I

am a data scientist working for a GAFA company and

those companies really emphasize on using data to improve their products

So this is my take on what is data science

Before data science, we popularized the term data mining in an article called from data mining to knowledge discovery in databases in

1996 in which it referred to the overall process of discovering useful information from data

In 2001, William S. Cleveland wanted to bring data mining to another level

He did that by combining computer science with data mining

He made statistics a lot more technical which he believed would expand the possibilities of data mining and produce a powerful force for innovation

Now you can take advantage of compute power for statistics and he called this combo data science. Around this time

this is also when web 2.0 emerged where websites are no longer just a digital pamphlet, but a medium for a shared experience

Facebook in 2004 and YouTube in 2005. We can now interact with these web sites

meaning we can contribute post comment like upload share

leaving our footprint in the digital landscape we call Internet and help create and shape the ecosystem

we now know and love today. And guess what?

That's a lot of data so much data, it became too much to handle using traditional technologies. So we call this Big Data.

That opened a world of possibilities in finding insights using data

But it also meant that the simplest questions require sophisticated data infrastructure just to support the handling of the data

We needed parallel computing technology like MapReduce, Hadoop, and Spark

2010 sparked the rise of data science to support the needs of the businesses to draw insights from their massive unstructured data sets

So then the journal of data science described data science as almost everything that has something to do with data

Collecting analyzing modeling. Yet the most important part is its applications. All sorts of applications.

Yes, all sorts of applications like machine learning

So in 2010 with the new abundance of data

it made it possible to train machines with a data-driven approach

rather than a knowledge driven approach. All the theoretical papers about recurring neural networks support vector machines became feasible

Something that can change the way we live and how we experience things in the world

Deep learning is no longer an academic concept in these thesis paper

It became a tangible useful class of machine learning that would affect our everyday lives

So machine learning and AI dominated the media overshadowing every other aspect of data science

experimentation, ... And skills we traditionally called business intelligence

So now the general public think of data science as

researchers focused on machine learning and AI but the industry is hiring data scientists as analysts

The reason for the misalignment is that yes, most of these data scientists can probably work on more technical problems

but big companies like Google Facebook Netflix have so many low-hanging fruits to improve their products that they don't require any

statistical knowledge to find these impacts in their analysis

Being a good data scientist isn't about how advanced your models are

It's about how much impact you can have with your work. You're not a data cruncher. You're a problem solver

You're strategists. Companies will give you the most ambiguous and hard problems. And we expect you to guide the company to the right direction

Ok, now I want to conclude with real-life examples of data science jobs in Silicon Valley

But first I have to print some charts. So let's go do that

(conversation not directly related to the topic)

So this is a very useful chart that tells you the needs of data science. Now, it's pretty obvious

but sometimes we kind of forget about it now

At the bottom of the pyramid we have collect you obviously have to collect some sort of data to be able to use that data

So collect storing transforming all of these data engineering effort is pretty important and it's actu-

It's actually quite captured pretty well in media because of big data we talked about how difficult it is to manage all this data

We talked about parallel computing which means like Hadoop and Spark

Stuff like that. We know about this. Now the thing that's less known is the stuff in between which is right here

Surprisingly this is actually one of the most important things for companies because you're trying to tell the company

what to do with your product. So what do I mean by that? So I'm an analytics that tells you

using the data what kind of insights can tell me what are happening to my users and then metrics this is important because

You know, these metrics will tell you if you're successful or not. And then also, you know a be testing of course

Experimentation that allows you to know, which product versions are the best

So these things are actually really important but they're not so covered in media. What's covered in media

is this part. AI, deep learning. We've heard it on and on about it, you know

But when you think about it for a company, for the industry,

It's actually not the highest priority or at least it's not the thing that yields the most result for the lowest amount of effort

That's why AI deep learning is on top of the hierarchy of needs and these things may be testing analytics

they're actually way more important for industry

so that's why we're hiring a lot of data scientists that does that. So what do data scientists actually do?

Well that depends on the company because of them as of the size

So for a start-up you kind of lack resources

So you can only kind of have one DS. So that one data scientist

he has to do everything. So you might be seeing all all this

being data scientists. Maybe you won't be doing AI or deep learning because that's not a priority right now

But you might be doing all of these. You have to set up the whole data infrastructure

You might even have to write some software code to add logging and then you have to do the analytics

yourself, then you have to build the metrics yourself, and you have to do A/B testing yourself. That's why

for startups if they need a data scientist this whole thing is

data science, so that means you have to do everything. But let's look at medium-sized companies. Now, finally

they have a lot more resources. They can separate the data engineers and the data scientists

So usually in collection, this is probably software engineering. And

then here, you're gonna have data engineers doing this. And then depending if you're medium-sized company does a lot of

recommendation models or stuff that requires AI, then DS will do all these

Right. So as a data scientist, you have to be a lot more technical

That's why they only hire people with PhDs or masters because they want you to be able to do the more complicated things

you probably have a lot more money and then you can spend it more on employees

So you can have a lot of different employees working on different things. That way