Table Extensions for Data Science and more : New in Tableau 2022.3 | Data Science in Tableau

0:00Hey, it's Tim here. In today's video, we

0:01are looking at table extensions. And for

0:03this video, we're doing something very,

0:05very different. The truth is, I don't know

0:07much about table extensions. And in fact, I

0:09don't even know anything about analytical

0:11extensions, apart from how to set them up.

0:13So what I did is I sounded the horn to the

0:15Tableau community, and a whole bunch of

0:17people reached out.

0:18Charles Laporte was one of them. And what

0:21he did is he walked me through how to set

0:22up table extensions, including an

0:24analytical extension and how it all works.

0:27We did a bit of a comparison to how they're

0:29different. But then the main thing we also

0:31did is we talked about some use cases. As

0:34of everything is timestamped below. So if

0:35you want to skip ahead, because you know

0:37something or you just want to sort of see

0:38the most interesting section of the video,

0:40then by all means use that. We're going to

0:42start off the video with Charles Laporte

0:43introducing himself. But as ever, if you've

0:45got any questions, leave them in the

0:46comment below. I'm sure me or Charles will

0:48probably reply

0:48to you. And yeah, as always, let's get

0:51stuck in.

0:51Okay, so my name is Charles Laporte. I am

0:55working for a company called Bistory. We

0:58are a Tableau partner, and I'm an analytics

1:01consultant. So my day to day job is going

1:04to Tableau and help people get the best out

1:07of their Tableau experience as well. And

1:11that goes from training, but mostly support

1:15and dashboarding.

1:16And I love to work in Tableau. And I love

1:19to work with other programming language

1:22like Python and HAR. So Tableau extension

1:25is really something I was looking forward

1:27to.

1:27Yeah, I guess like you've used, I assume

1:30you've used analytics extensions already,

1:33it might be worth for everyone's benefit,

1:35just explaining the difference between the

1:38way analytics extensions worked before with

1:42calculations and why it's different with

1:44Tableau extensions.

1:46Yeah. So before, when Tableau released

1:53their first analytics extension, you could

2:01already connect to Python using the Tap

2:08R serve, which is actually a library in R

2:12creating this R server where you can push

2:14code and then it comes back. And you had to

2:16use, like you said, a calculation. So if

2:21you're in Tableau, you might remember that

2:23you have to script underscore something

2:26calculation, which are table calculation

2:29that allows you to send some code towards

2:32an analytic extension, and then it goes

2:34back as data.

2:37The good side of it is that you could do

2:40advanced analytics. So for instance, you

2:43could deploy a very complex model where,

2:46for instance, you analyze data. And I think

2:50Tableau had a very good example with cancer

2:52data. So I think it was breast cancer data.

2:57And you could, through parameter, you could

3:00enter some data, and it would return the

3:04probability of cancer or not.

3:07Right. And that was through deploy model on

3:10Tap by the downside. The downside of this

3:13is that we are talking about table

3:15calculations. So if you remember correctly,

3:19table calculations need to be compute

3:22following a certain scope and certain

3:23direction. So it needs to go on desegreg

3:27ated data. And that's why, for instance, I

3:30if I request my, my model to return the p

3:35value, it would return the p value.

3:36It would return the p value for each of my

3:36points in my view, which sometimes is not

3:36right. Not the not the best. Okay, so that

3:36's the start of an actually beautiful. Yeah,

3:36that's the start of the analytics extension

3:36. And then it evolved straight into Tableau

3:36with more calculations. I don't know if you

3:36know about the model calculations that

3:36actually also allows you to kind of

4:06get more information about your model. And

4:09then they released the analytics extension

4:13API.

4:13Right.

4:15Which actually allows you to use almost any

4:18programming language to create models and

4:21advanced analytics.

4:24I got that. So they they they built Tap Pi,

4:28there's the is it is our part of MATLAB or

4:32is MATLAB completely separate? MATLAB is

4:36completely separate. So HAR is the

4:37statistical programming language that was

4:39founded a couple of years ago already by a

4:44teacher in New Zealand. And now it has a

4:46pretty big community within the statistical

4:49world.

4:50I don't know if you know a company that

4:52previously was called R studio. They

4:55recently changed name to posits. Well, they

4:59are pretty strong in in the R community.

5:03And they actually the main IDE for R. And

5:06now they are trying to integrate Python

5:08users as well. Because yeah, you I don't

5:11know if you are aware of this Python versus

5:13our fights going on.

5:15Yeah. Yeah, there's always a fight between

5:18it. Someone will say our Python versus I

5:21don't know, and any any other program,

5:24right?

5:24Yeah. If you go into the programmer wall,

5:27people will say C sharp is better than

5:29Python. And then Python user will say

5:32Python is better than our and then our user

5:34will say, Look, I'm using R and Python. So

5:37I don't care.

5:39I think something is that's true with both

5:41R and Python is that they do rely on

5:43modules. So these are sort of packages that

5:45have been built specifically for the

5:48languages to make it easy to do certain

5:50things, just using a function, right?

5:53Exactly. So, as you go along with

5:55programming, you realize that at some point

5:58, you are repeating some tasks. And that's

6:00where the package going Zendy, because

6:03people just said, Okay, instead of doing

6:06your linear regression over and over, and

6:08then do all those steps, actually, we can

6:11create one function that just take up your

6:13data frame, and then returns you all the

6:15results.

6:18Right. And actually, there is a rule of

6:20thumb for that, that we could somehow apply

6:22to Tableau as well, like if you're doing

6:24things three times, or more than three

6:26times in your code, it's time to write a

6:28function.

6:30Right. Yeah, that's actually true, right?

6:33We, we, we try and write calculations, like

6:35templates, sometimes we try and think I've

6:38seen a few people write, you know, work

6:40books with pre done calculations, right? And

6:43I think many features in the Tableau ideas

6:45page are gonna have my snippets saved

6:49somewhere, right? So that you don't have to

6:51keep writing the same, if this then that

6:54calculation, right? So yeah,

6:56I would love that. I would love that for

6:59the day calculations, because many times

7:02you get this request on, oh, yeah, I want

7:04to have this year to date calculation,

7:06depending on a parameter, and then you

7:08start to start to mess up your mind around

7:11her.

7:13Now, at this point, it's probably worth

7:15explaining the term extension, because

7:17Tableau has recently started using terms

7:19everywhere. So let's start right from the

7:21top. Anytime you hear the term extension in

7:24Tableau, it simply means a capability that

7:26goes beyond the native capability of Table

7:28au itself. So you might have heard of

7:30dashboard extensions, they're designed to

7:32help you build functionality that Tableau

7:34don't build themselves and add it to your

7:36dashboard. You might also have had of

7:38analytic extensions, those are essentially

7:40capabilities that allow you to add another

7:43analytical component, in this case, our

7:45Python MATLAB or any other analytical

7:47service. And what that does is allows Table

7:49au to send the calculations to that service.

7:51That's essentially how analytics extensions

7:53work. Now table extensions require

7:56analytical extensions. But the difference

7:59with the table extension, and it changes

8:00the way the data model and the table works,

8:03when you bring a piece of data in it

8:04essentially happens at the point of

8:06connection. So it's an extension because it

8:09changes the way you connect to data allows

8:11you to go beyond what Tableau can already

8:12connect

8:13to, either by calling an API or by

8:15specifically calling a particular

8:16analytical function from an analytics

8:18extension. But the really powerful thing

8:21here is it does it at the point of

8:23connection and can also append it or add it

8:25to a data model for a live connection,

8:27which means it works in real time alongside

8:29your data set. So you can in real time call

8:32API's as your data changes. And so it's a

8:34really, really powerful capability. I

8:36thought it was just worth clarifying that

8:37before we carry on. So you're absolutely

8:39clear about all the different extension

8:41types we have in Tableau. Okay, let's hope

8:43back in.

8:43Yeah, to me, I think the offering of

8:46advanced analytics in Tableau doesn't

8:48really target the people that will make

8:51machine learning in production. I think we

8:54are touching to people that are doing ad

8:57hoc analysis within through Tableau. And

9:00then at some point, they are thinking, Okay

9:02, well, I want to confirm that hypothesis

9:04with a statistical test, which I cannot do

9:07directly in Tableau. So I can get help from

9:10from our or Python.

9:13And the good advantage of table extension

9:16is that before, like we said, it would

9:19return from a table calculation. And right

9:23now, you can actually put it back into a

9:26table, a data frame. And you can either

9:29choose to put it back as a new field in

9:32your data or a new table. So that's a great

9:36, great addition.

9:39Good work. Okay, so, um, do you have an

9:42example we can kind of walk through or like

9:45, what are we looking at? What is this?

9:48Well, yeah, this is already a bridge too

9:51far. So I'm going to start clean and create

9:55a new data source. And actually, I'm going

9:59to pick a text file. And in this case, I

10:02have a huge folder with a lot of data sets.

10:06And I'm going to pick my penguin data sets.

10:10It's actually something if you know Bistory

10:12, we love penguins. And I don't know if you

10:17know

10:17that. Well, no, you know, and there are

10:20funny stories about it. So penguins, pengu

10:25in data sets.

10:26Do you have a mascot?

10:27Yeah, we have a mascot actually for for

10:29that as well. It doesn't show off in the

10:32outside world. But internally, we are very

10:35proud of it.

10:36Inside, inside the lid as a mascot. All

10:38right. Okay.

10:38Yeah. So here we have our penguin data sets

10:43. So it's a famous data set. It's actually

10:48very famous in the R community. It's about

10:50polymer penguins. So a study made about

10:52penguins in the South Pole, sorry. So you

10:58have different species from different

11:00islands. And it's actually all Adélie

11:02penguins. And they collect the data about

11:08their

11:08pill and their flippers and their body mass

11:10. And what we want to do here, we want to

11:15create a linear model that could help us to

11:18understand the relationship between the fl

11:20ipper length and the body mass. Of course,

11:24of course, what you can do is just go into

11:26Tableau, create a scatterplot, where we

11:31have our flipper length and body mass.

11:34And yeah, I need to desegregate. So there

11:40we go. We have our scatterplot. I can just

11:43put a trend line and that's it. And now we

11:47know, okay, we have a pretty confident, we

11:50have quite high R squared. So that tells us

11:53that or the model that we have built now

11:55with this trend line accounts for 75% of

12:00the variance that we see in the data.

12:02So pretty good. And if you want to look at,

12:07if you want to speak about correlation,

12:09well, you can clearly see that there is a

12:11positive correlation here with this data.

12:14But what if we really want to check the

12:17assumption of this linear model? So I don't

12:21know if you know, but you have actually

12:23four assumptions behind linear regression.

12:28And one is very important is we shouldn't

12:32see any trends between residual and fitted

12:36values. So it should be a random dot cloud.

12:45The residual, so the difference between

12:49your actual points and the fitted points on

12:53the line, that distribution should be

12:58normal. And those kinds of assumptions, you

12:59cannot directly check in Tableau. Of course

13:00, you can make hypothesis. I could easily

13:02calculate the difference between my actual

13:05value and the fitted value of the model

13:08because I have the equation here.

13:10But yeah, that's some extra steps. And that

13:15's where Tableau extension comes in very

13:17handy. So let's go back to our data source

13:21page. And as you may have seen in the

13:23latest version of Tableau, we have table

13:26extensions.

13:28So I'm going to remove my Penguin table and

13:31I'm going to pull up table extensions. And

13:34the first thing I'm going to do is follow

13:37Tableau's instruction. I'm going to drag a

13:41table here. So I'm going to look for my

13:46Penguin file, put it here.

13:53One really important point, set your

13:55analytics connection. So if you go to the

13:59up menu, setting performance, manage

14:02analytic extension. I already set it up

14:05front.

14:06So through my command line, I open R and I

14:11use this command, colon colon rserv parent

14:16hesis. And now my rserv is running and I can

14:22test it here. It should return a

14:25successfully connection. Yes. So I can save

14:28and close this.

14:29Is that what you've got in the command line

14:32? Is it something you trigger or how did you

14:34start that up?

14:35Yeah. So rserv is a specific library of R.

14:40So I could have started R studio. So I have

14:44R studio here. I could have started R

14:46studio and load the library and then run

14:48the command rserv. Usually what I do, I

14:51just trigger it through the command line.

14:53You can also, if you want, trigger it in

14:57other ways. I know that I'm not using the

15:01most convenient way, but for what I'm doing

15:05with R. Yeah, exactly. Exactly.

15:07There's no right way of using anything. I

15:09always tell people, you're one of these

15:11people that says use it this way. Like you

15:14're not using it properly.

15:16Exactly. Well, there are some areas where I

15:19can improve that's for sure. But for now,

15:22that's good for me. Okay. Start again.

15:26Table extensions, we drag the penguins

15:29table. And as you see here, we have an

15:32input and output table. So our input table

15:36is our penguin.txt file.

15:38Our input table is now empty because we

15:41need to enter a script. And the script is

15:44kind of the black box right now because we

15:47don't have many information about the

15:50script.

15:50And to be honest, when I started to use the

15:53table extension, I was like, okay, script

15:56looks like a bit how we need to use a table

15:59analytics extension in Tableau prep.

16:02So I was starting to write a function. So

16:05if you remember correctly in prep, you need

16:07to write two functions, one to actually

16:11transform your data through Python and R

16:13and another one to get the output schema,

16:17for instance, if you want to have another

16:20table in another form.

16:21So that's what I did. And I ran into many

16:24errors. So I was like, Hmm, that's not

16:27correct. So went back to the actually the

16:29new features video of Tableau.

16:31And actually noticed something very

16:33interesting. They were using this dot arg

16:37and that's specifically for hard or

16:40underscore arg.

16:42And if you remember correctly in the script

16:46calculation, that's so you could call your

16:50field from your data source.

16:53So starting from there, I said, okay, well,

16:57let's just type dot arg one and see what

17:01happens. So if I click on apply, nobody

17:02should trigger something. Yes. Processing

17:05the request.

17:07And you will see that actually Tableau

17:09normally will return my whole Penguin table

17:12. So that's one way to interact. So if I go

17:16to output table, no, I have my data.

17:19It returns everything. Yeah, it returns

17:23everything. When you set arg one and then

17:24you ran the script, what happened is arg

17:27one is just a variable, I assume, right?

17:30Yeah. So our team president is we are

17:35telling her arg one is my table within my

17:40table extension. So I'm telling her look,

17:46print penguin dot txt. Exactly. That's what

17:52it's happening.

17:53And it's grabbing the data. Exactly.

17:58Processing, doing nothing to it. Exactly.

18:00It's just grabbing. So for instance, for

18:02now, if I'm saying okay, I want to return a

18:06specific column in arg if you use the

18:10dollar sign next to your data, and I do

18:14body mass in gram, and I click apply.

18:19Normally it should return only one column.

18:26I never tried it. So.

18:29Okay. Living life from the edge. Okay.

18:33Doesn't work. Okay. And I know why. Because

18:38through the many steps I made towards a

18:40good result, I noticed that actually Table

18:44au doesn't send the data to WatchR or Python

18:48.

18:48As a data frame as a table, it does send

18:51the data as a dictionary. So really, I don

18:55't know if you're familiar with dictionary

18:59in Python.

19:00Not fully. Okay. Explain it to me, maybe.

19:03Yeah. So basically a dictionary is, and I'm

19:06going to write the syntax here. I don't

19:09know if it's very clear, but.

19:12It takes up the name of your column. So let

19:15's imagine we have our species colon. And

19:19then within that specific for that specific

19:23key. Oh, like an array. Yeah. Like an array

19:26for that specific key, then you have some

19:27values. So then you have Adelie. And then

19:31in this case, it repeats itself for many,

19:35many times.

19:37Because different ways. Yeah. So it's

19:39similar to like a JSON array or something

19:42which uses slightly different notation, but

19:45it's the same sort of structure.

19:47Yeah, it's always the same kind of

19:49structure. So you have a key and then you

19:53have values behind it.

19:55And actually that's kind of tricky because,

19:59uh, in Python, it's quite okay because

20:02usually data is interpreted as a list and

20:05dictionary and list are pretty close to

20:07each other.

20:08So it's easy to work with in R they

20:11actually are using a class of, uh, or data

20:14type that is not used anymore called the

20:17parallelists.

20:18So dictionary in R are not very common

20:22because R most of the time works with

20:26matrix or with data frame.

20:29So I'm going to switch back to my actual,

20:32uh, example where I already did more code

20:36and you see here, uh, that I, so I created

20:42a linear model.

20:44And for that, I'm using the LM function,

20:47which stands for a linear model. And if you

20:49look at the data, I actually convert it

20:52first.

20:53So what I do is I first convert my R one to

20:56a list because yeah, R doesn't interpret

20:59for the LM function.

21:01You need to use a data frame or list. So I

21:03first use it, uh, covered as a list.

21:06And I decided to convert as a data frame

21:09because you know, the steps I made, I

21:11realized that that might be the most

21:14convenient.

21:15And then I, uh, saved everything under my

21:22model.

21:22And from there, uh, I cannot use the model

21:28object directly because it doesn't return a

21:34data frame. It returns, uh, an object of

21:36class model, uh, which is not interpretable

21:37directly in Tableau.

21:37And I'm actually loading another library

21:40called broom and the broom library actually

21:43allows you to return two things.

21:45So you have two main functions of the

21:47augment function, which allows you to

21:49retrieve the two fields that the fields

21:51that you are using in your calculation.

21:53So in this case, I'm trying to predict body

21:57mass, uh, from flipper length and gender or

22:02sex.

22:03And the augment will return me a data from

22:05where I have those three columns plus

22:08defeated values, plus the residual.

22:10So really all the information that you can

22:13see here, all the information I could get

22:16from my model, which will be very useful

22:18when I want to build a visualization based

22:22on my model.

22:24Then if we go a little bit down, I added

22:28some more code to retrieve something else.

22:32I will come to that later. And in the end,

22:36I'm actually returning a data frame.

22:40Uh, and that's something I asked to Tableau

22:43as well, because I it's a bit cryptic to me

22:45, but somehow when you're working with

22:48Python, do you need to go back to a

22:50dictionary, but in R sometimes data

22:52framework directly.

22:54So it's something I, um, I dropped the

22:57question at Tableau like, okay, what's

23:00going on there? Yeah.

23:02They said documentation was, was coming

23:04soon enough. So I was going to say the

23:06documentation is kind of tough in this, in

23:08this space, isn't it? Cause, uh, you have

23:11to work on assumptions and little snippets

23:14from videos, right?

23:16Yes. Well, to be honest, I think, I think

23:19if you were in the, if you're part of the

23:22data dev community, I think they hosted a

23:25couple of sessions about it, but besides

23:28that, if you read the Tableau help, it's

23:31pretty, pretty empty at the moment.

23:34And I think for people to engage in the

23:37community.

23:38Yeah, I think the best tip would be to

23:40engage your community and, and honestly, uh

23:42, that's what I've been doing the last

23:45couple of days. I've seen some people

23:47asking around like, okay, what, what do I

23:49need to do with this?

23:50Because I want to use it. And I've seen

23:52some example from Tableau where, um, they

23:54have actually pretty cool example.

23:57And like, I think in the feature video,

24:00they are actually using a sentiment

24:03analysis for tweets, which is super great

24:07to do.

24:08But if you don't know the syntax of the

24:10script, you're quite blocked.

24:13So if we go back to my example with a Bingo

24:20model, so now I have this data frame with,

24:22uh, so if I go to the output with all my,

24:23uh, statistics, statistics around my model.

24:25So I have my data here and then I have my

24:28residuals and I can start actually already

24:33to build some visualization.

24:36So I did here. So I created this scatter

24:40plot using body mass and flipper length,

24:43and then I created dual access with a

24:45fitted value so I could return those, uh,

24:48trend lines.

24:52That's very cool. The value of this is,

24:56yeah, go on.

24:57Uh, yeah. So what I was going to say,

24:59actually at this point, what you see here,

25:02you could already do it with the previous

25:04version of the analytics extension in Table

25:05au.

25:06So using the script calculation, I could

25:10easily return a fitted, uh, column where I

25:14could, that I could use to make a scatter

25:17plot and through a trend line on it.

25:20So no, it's just another way. It's just

25:23instead of having a one array of data, I

25:25have actually a table behind it.

25:28Is this faster because it's not having to

25:30do it every time it loads the chart. So it

25:33's pre-computed in the data set. Is that a

25:36performance benefit?

25:38That's a question I'm actually asking

25:41myself every time I'm using it because the

25:45first time I tried it was super, super slow

25:49and I used it through R serve on my local

25:52machine.

25:53I use it through top on my local machine

25:55and I use it as well with a top by instance

25:58that I have on Heroku.

26:00And yeah, well, performance is really

26:03variable. Like sometimes with the Heroku

26:07instance, it was super fast.

26:09Sometimes with the R local R serve, it went

26:12super slow.

26:14The thing I notice is, uh, and you didn't

26:16see it because I opened the work before is

26:19if your analytics extension is not running,

26:22the work group will open blank.

26:24And even if you trigger the analytics

26:27section and you go through the help menu to

26:31set it up, it will remain blank.

26:33So you really need to go back to the data

26:35source, click apply on the script, and then

26:38the data will appear.

26:40Well, I guess, yeah, that's pretty, pretty

26:43new as well for, for Tableau.

26:46Yeah, they'll probably need to work on that

26:48kind of improvement because you would

26:49expect that behavior to be affected by the

26:51pause and the play button that we have in

26:53the interface at the top, rather than

26:56having to go back to the data connection.

26:58Yes.

26:59Well, also one other important point is

27:01that the data that you retrieve from Python

27:04and R cannot be extracted.

27:06So if you look at my data source here, I

27:08actually cannot use an extract.

27:11So it's always all alive. So it's really

27:13Tableau sending the data, retrieving the

27:16data.

27:17And if the analytics extension breaks, well

27:20, it breaks the workbook basically.

27:23High stakes.

27:26Exactly.

27:27But yeah, again, I think this is not meant

27:32for, uh, yeah, machine learning and

27:34production.

27:35And although we have seen some great

27:37example with Tableau in the past on, yeah,

27:39you can actually deploy your model.

27:42I don't think any companies or I believe no

27:44companies will ever use that in production

27:47and say, look, we put our model on a

27:51instance and that's what we are using.

27:55But for data and I was going to say for

27:57data prep, I'm wondering, I don't know how

28:00this works with Tableau prep, but for data

28:03prep where you can potentially bake some of

28:06those results in, it might be useful.

28:10But in that case, you're probably doing it

28:13outside of Tableau or in prep or in

28:15whatever other tool you're using anyway.

28:18Yeah. And actually that's one of the use

28:21case I wanted to put forward is so we have

28:24this statistical, all the statistical stuff

28:27and machine learning where you can build

28:29your model, but you could also do advanced

28:31data prep through Python and HAR.

28:34It's super easy through Python to pivot and

28:38you can actually have a better pivot than

28:40the built in pivot of Tableau Desktop.

28:44So things that are possible, but again,

28:47yeah, you need to balance your choice

28:50depending on, yeah, well, am I doing a one?

28:56Am I doing this once or am I going to do

28:59this and launch it into production?

29:02Because then, yeah, I wouldn't do it in

29:06Tableau.

29:07But let's say you want to make a Sankey

29:10chart and you need to union your data with

29:13himself and you don't want to use the data

29:15model for some XY reason and you want to

29:18use the analytics extension.

29:20Well, that's a good use case as well.

29:24And while we are speaking about use case, I

29:28think the one that people are most looking

29:32forward to is actually get API data.

29:36So if we go back to this, so our script is

29:40not bound to what's in the data model.

29:44So I can actually call something, an object

29:47that is outside of my Tableau environment,

29:51my Tableau workbook.

29:53And one way to show this is let's go back

29:58to this, to this other data source.

30:01And it's actually something I tested out of

30:05. Yeah, I was a bit lost in this table

30:07extension.

30:08And at some point I was like, okay, let's

30:10just type Iris. Iris is one of the built in

30:13data set of R and I clicked apply.

30:16I was like, yeah, you never know. Maybe it

30:18could work.

30:20And actually it works. So you can call

30:25other data sets directly.

30:27Yes. I guess this is an amazing opportunity

30:32for APIs.

30:33Yes. An example I've seen the Tableau

30:37shared with me was calling the YouTube API

30:40to go get statistics from YouTube.

30:44Another one was Google search. So you've

30:46got some terms in your data set and you

30:48want to go search the most commonly

30:50searched related terms for marketing

30:52purposes.

30:53So you can then build a dashboard on which

30:56terms to go and bid for and how much they

30:58cost.

30:59That kind of stuff would be phenomenal

31:01because at the moment there's still not an

31:04easy way to do that.

31:06Right. And this would lend itself, I guess,

31:11to would this be an ad hoc capability?

31:13Wouldn't be something you'd put in

31:14production, but potentially for someone who

31:16's more of a data scientist using a lot of

31:19tools,

31:20this could be a nice way for them to sort

31:22of build a picture to share with people.

31:24Yeah, I think what you can do is quickly

31:28build a proof of concept through this, like

31:31saying, oh, I want to retrieve data from an

31:33API and I want to link it with some data

31:37that we have in our database.

31:39I can quickly do it through Tableau and

31:41through the analytics extension without

31:44starting to create this big Jupyter

31:46notebook where it's kind of iffy to connect

31:49the database.

31:50I don't know if you've tried to connect the

31:52database through R. I don't I don't like it

31:54.

31:55That's why I like Tableau. It's just, oh, I

31:57need to connect to a database. Well, it's

31:59just new data source.

32:01I just pick it in, pick it, pick my DB in

32:04the list and that's it. Well, in R I have

32:07to start to type a string, a connection

32:11string and then username, password.

32:14So to me, this is the best of the two,

32:16especially like in case of you want to

32:18build a quick proof of concept, you want to

32:20prove something quickly by,

32:22oh, I want to link the database or data

32:25sets with my API data that goes super quick

32:29.

32:29And certainly I think API is something that

32:32I hear more and more also from clients, but

32:35also from other people within the community

32:37where they are saying, well, we are missing

32:41this API connector in Tableau.

32:43This connector, yeah.

32:45And of course you can use.

32:46Everyone thinks that connector is special.

32:48Yeah, yeah, exactly.

32:50And you have, of course, the Web Data

32:52Connector 3.0, which allows you now to

32:56actually, so before you had to use a web

32:58server to host the connector, apparently,

33:01no, you don't even need the web server.

33:03So I think we are getting closer to this

33:06API connector, but in the meantime, yeah,

33:10we are working.

33:11This is a nice way.

33:12And I don't have an example right now

33:14because I actually have one example.

33:17Sorry. So if I go to my Datadev instance, I

33:25made this useless fact dashboard.

33:28So you have a useless fact API, which is a

33:32very simple API that returns a fact, a new

33:36fact every time you refresh it.

33:41And it's very easy to interpret. So with

33:44the minimum Python code I could write, I

33:48could retrieve it, and I decided to make a

33:51dashboard.

33:52I forgot my site, of course.

33:56This happens to me every time.

33:58Yes, yes.

33:59I programmed a shortcut to remember my

34:03online instance URL.

34:06That's the thing. So when you're subscribed

34:09to the Datadev community, you get a Tableau

34:13site on Tableau Cloud.

34:15And I chose this very long name, bstory-

34:18charles-lapport-dev.

34:20And then they added this dev with a number

34:24and I cannot remember.

34:27So let's go to my test.

34:30My naming convention is really bad. I'm

34:32sorry for that.

34:34This is called test and then you have sheet

34:36.

34:36Everything is attached until it's final,

34:38final, final, final.

34:40Exactly.

34:43And in this case, what I did, I used the

34:45Tableau extension directly on Tableau Cloud

34:47because that's where it was.

34:49It was actually deployed before the release

34:51of 2020.3.

34:53So you could already use the Tableau

34:54extension in Tableau Cloud. And that's what

34:56I did.

34:57And normally a useless fact should appear

35:01in a couple of seconds. Oh, there we go.

35:04There's just kisses are called that because

35:05the machine that makes them looks like it's

35:07kissing the conveyor belt.

35:09Useless fact.

35:11I did not know that.

35:14That's a great fact.

35:18Well, it's a great, it's a great API too.

35:21But actually, I was starting to look into

35:25APIs and some person was sharing, OK, those

35:28are the easiest API you can use.

35:30And this one was like, yeah, it's super

35:31easy.

35:32It just returns the facts as text timestamp

35:35and that's it.

35:36And it's very easy to interpret.

35:38And actually, if we look into the workbook

35:43itself.

35:44So let's use a little bit web authoring.

35:47Now we go to the data source, you will see

35:53that my Python code is ridiculously short.

36:01So if I go into this, you see it all in one

36:10, two, three, seven lines. So importing the

36:18important libraries to the request to read

36:18API data, pandas to do the Gson parsing.

36:21And basically just, OK, I have this useless

36:23fact.

36:24And actually, I think it works in my

36:26browser as well.

36:28So if we want to have another useless fact.

36:32There we go. What does it say? Hot water is

36:35heavier than cold water.

36:38You always learn something with this.

36:41And you see this.

36:44I did not know that. Why would it be

36:46heavier?

36:48I have no idea. Now I'm getting into the

36:50science of it.

36:51This is a completely distracting thing.

36:58Yeah, but there's a thing with this API as

36:58a side note.

36:58I'm also a person that can get attracted

37:01out of things I read like.

37:03And those kind of things.

37:05I'm triggered to look through Google like

37:08why is hot water?

37:10Explain why. Yeah, explain to me why.

37:16Oh, God. So this is quite nice. This is a

37:19Python example.

37:21The things you're importing. Do those have

37:24to exist in your tab pi instance?

37:26Or are you just calling libraries from the

37:29Python?

37:30My example here is I use something called

37:33Homebrew on my Mac.

37:34And with Homebrew, you can just call

37:36libraries from wherever and then it imports

37:38them and installs them.

37:39What's going on here?

37:42Well, so I'm actually using my data dev

37:47papa instance on Heroku. So you can

37:50actually deploy it yourself.

37:52I don't know the limitations, actually.

37:56It was quite surprising that they would

37:58allow me to deploy it for free.

38:00But afterwards, I discovered that you could

38:02also have a free Postgres database on Her

38:06oku.

38:06That's changed recently.

38:09But basically, that's there. And I think

38:12those libraries were already in the tab pi

38:15instance.

38:17Yes, they're already included.

38:20I think it is possible to add libraries to

38:24it in the settings.

38:25I think there's a settings file where you

38:27can add, you can tell it what libraries to

38:29include.

38:30Well, if we go to the tab pi instance, I

38:40can quickly...

38:40Yeah, of course, I need to enter a password

38:48.

38:48I'll blow this out, don't worry.

38:52Whenever I edit videos, I'm paranoid

38:55because the amount of videos I make and the

38:59URLs and logins I share,

39:01I'm going to get dosed by someone. Whenever

39:04I edit video, I'm religious about blurring

39:07out user names.

39:08Yeah, but you're right to do so because we

39:12live in a crazy world and people like to

39:15act.

39:17Yeah, like a social engineer even just with

39:21me.

39:22Yeah, and also people like to do that. That

39:24's what we do now with the table extensions.

39:26That's why one of my colleagues said you

39:28are really hacking the feature.

39:30There is no documentation whatsoever, but

39:32you have spent some time just to find out

39:35how to use it.

39:36Just to find where it works. True, true.

39:39I was saying, yeah, if you look at the...

39:42That's a tab pi instance running on Heroku

39:44and you see I don't have any deployed

39:46models.

39:47That's also something you can do.

39:49If you work from Jupyter notebook for

39:51Python or even through VS code, what you

39:54can do,

39:55you have a specific function within the tab

39:58pi package that allows you to deploy

40:00something.

40:01Even if the library doesn't exist right

40:03away in your tab pi instance, you can

40:05actually deploy it.

40:07And that is something I never used in the

40:10past.

40:11But basically I could just deploy a

40:14function with my useless fact model to make

40:18it easier for people just here to say,

40:20"Okay, use this." And then it would

40:24retrieve a useless fact anytime I do it.

40:28One important point when you are calling

40:30API and you may have noticed it is that I

40:33don't have data here.

40:35So that's a bit the weird side effect of it

40:42.

40:42And you have the same in prep is you need

40:44to create a connection to a random file.

40:46Here I picked up the global superstore

40:49Excel file.

40:50Just to have this, okay, Tableau has a

40:54connection and I can use my table extension

40:57.

40:58Okay. And it's probably worth talking a bit

41:03more here because in our example,

41:07you just connected directly to the table

41:10extension.

41:11But am I correct in thinking you can still

41:13use a table extension as part of a data

41:15model?

41:16So you can relate whatever you're computing

41:19in the table extension to a data model,

41:23which then makes it part of your sort of gr

41:25ander picture.

41:26Yeah, exactly. Exactly.

41:29That's amazing.

41:30If I go back here, I could close this.

41:33So you see my table extension leaves on the

41:36logical layer and I could just take my pngn

41:39file

41:39and create a relationship between the two.

41:42Of course, here it will work because I

41:44loaded the iris table.

41:46It's going to try and create like a...

41:48Yeah. Interesting.

41:50So it's super powerful then.

41:53I think this is one of those things where

41:55maybe we're just going to continue to see

41:57people doing things with it.

41:59And every time it will be new because the

42:02effort is really on the organization or the

42:05individual to sort of build their own

42:08solution.

42:09So this is one of those feature sets where

42:11Tableau are giving, I guess, developers the

42:14tools to go and build these very bespoke

42:16solutions

42:17where it doesn't make sense for Tableau to

42:20be building a connector for penguins.txt,

42:23the connector for the Google API, all these

42:25small things.

42:26Yeah. And I think you know how Tableau has

42:31been working on leveraging AI on the

42:35platform.

42:37We have AskData, we have ExplainData, but

42:41still those remain data stories.

42:45And now we have DataGuide, which is not

42:47really AI, but help us to understand the

42:51full story.

42:52And I think this analytics extension still

42:55remains a little part of it.

42:57I hope that this table extension will

43:00actually make it grow a little bit bigger

43:03because the limitations that we had before

43:06with table calculations are not existing

43:09anymore with this table extension.

43:12Super cool. I'm still waiting for someone

43:17to... I don't know if you're aware, but in

43:18the preview of 2020.4, there's a new data

43:24role called image roles.

43:26And with this capability, I'm waiting for

43:30someone to write an API call to mid journey

43:34to create an image based on a description

43:37in their data set,

43:38and then load that image as an image role

43:41in Tableau. That will be the demo that I

43:43think goes down.

43:45Maybe you can build it. That will be the

43:46demo that will blow people's mind because

43:49that's just one of those sort of full

43:52circle platform sort of showcases, right?

43:55Yeah, yeah, indeed. Yeah, that should be...

44:01I'm thinking about it now. That should be

44:03possible. Something I wanted to do...

44:06It just depends. Yeah, go on, sorry.

44:10One thing I would like to have here is

44:14actually the ability somehow to integrate

44:19parameters because while I was working on

44:22this model,

44:24it wouldn't be great if actually people

44:26would be able to change this formula to

44:29whatever they want based on the parameter

44:32because right now it's pretty fixed.

44:35So if I'm the data scientist working on

44:37this workbook, it's fine. But let's say

44:39someone else wants to work the workbook.

44:41He needs to go to the data source, write

44:44this, run the...

44:47If we could just be in this dashboard I

44:50made and just, okay, I want to change the

44:57formula and not flipper length and sex, but

44:59I want only flipper length and bill length,

45:03for instance.

45:04That could work as well. Yeah. And then it

45:07reruns the model and boom, you've got it.

45:10Yeah. Interesting.

45:12And in terms of... I'm going to sort of go

45:16on a tangent here. Do you work with Snow

45:20flake a lot?

45:21I personally don't, but I did have a couple

45:26of demo of Snowflake.

45:29Yeah. So the area that I'm sort of

45:31interested in is Snowflake also talk a lot

45:34about Python and Snowpark, right?

45:36So the ability to run Python in the Snow

45:39flake platform. And so I wonder, like, is

45:44the benefit of doing it in the database?

45:46Is that it's done outside of your workbook

45:49and then your workbook can just call the

45:51results versus this, which is, I guess,

45:54more, as you say, like the prototype.

45:56You're just trying it out. You're trying to

45:58see the outcome. And then once you have the

45:59outcome, you move it further back into the

46:02stack so that it's done in a more

46:04operational, visible way that other tools

46:07and other analytics tools can also see.

46:09Yeah. To me, and that's why I think it's

46:13also why those analytics extensions remain

46:17quite small in the Tableau world is that at

46:21some point people realize that if you want

46:25to launch a model in production, you cannot

46:28do it directly in Tableau.

46:30You need to go up front in the web. Yeah,

46:34you need to do it before in the data prep

46:39step. And I think this integrate Python to

46:41Snowflake is a great addition to that

46:42because it limits the connections towards

46:47other services.

46:49And to be honest, I haven't looked into

46:52detail how they do it. But if it's

46:55integrated properly, then you really have

46:58an advantage.

47:00And I think that's what people are looking

47:03now on how to, how can I deploy a model and

47:06make sure it's working and have continuous

47:09improvements and continuous developments.

47:12And yeah, using Tableau extensions, the law

47:16doesn't allow this iteration, at least not

47:20as easy as it would be in a previous step.

47:24So the ideal solution then would be that

47:26when you connect to data, instead of having

47:29to like, copy and paste your code in or

47:32type your code in, being able to call like

47:34a library of code, I think it's called a

47:37notebook, right? So be able to call a

47:39notebook, where someone has already written

47:43code, it's visible to the organization.

47:45And if that code changes, your code changes

47:48as well. So there's a little bit of version

47:51tracking and capability there as well. So

47:54that that would be sort of a fully

47:56integrated sort of way of doing that.

47:59But, you know, it speaks to this idea, I

48:02think we've, we've lost track of this in

48:05the community that Tableau was always about

48:08exploration, as much as it is about

48:10building the dashboard and the final output

48:13.

48:13And this is a feature to me that I think,

48:16yeah, this is this is like giving you the

48:19tools to go out on an expedition and find

48:21the insight. And once you found it, you can

48:24think about how to bring that to more

48:26people in a much easier way. But it's it's

48:29a primarily exploratory toolset, right?

48:32Yeah, for sure. And linked to what you say,

48:35I think explained data is also a major

48:38changer in that game as well. Because

48:41before when you were going to explore it,

48:45to exploring your data, you had to combine

48:48yourself, oh, okay, I have this dimension,

48:52I want to see if there is a relationship

48:54between this dimension with that measure to

48:56other dimension, and it would take you

48:59quite some time.

49:00While now I can just take this point, click

49:04on the light bulb. And then here it says,

49:07Okay, some of body mass is high. Do you

49:10have an explanation for it? Well, actually,

49:14yeah, in this case, it doesn't have really

49:17have a great explanation. But at least you

49:20have leads towards a great explanation or

49:25potential explanation.

49:26Yes, exactly. Which is for me, a big game

49:29changer when you do data exploration. And

49:33in a business context, of course.

49:37Yeah, I think it's sorry, I think it's also

49:40important to take into account the

49:42knowledge that people have about their data

49:44. Because some people asked me, Yeah, but

49:47this is a bit rubbish, because it doesn't

49:50say anything. I said, Yes, but if you are

49:53someone with, with the business knowledge

49:56behind data, things can become very clear

50:00very quickly.

50:02Yeah. And it's, it's a really important

50:05step. This is I sort of see this as a

50:07painful step towards this feature getting

50:10better, because it's only by people using

50:13this initially and sort of iterating

50:15through it, you have to build it in public,

50:18you can't build it in private and then say

50:22that, because it doesn't get good that way

50:23these AI machine learning tools get better,

50:26as people use them, and they start

50:28collecting that sort of usage.

50:31And then they start to be able, you know,

50:33three, four years down, I mean, Google

50:35search, you know, we all now think Google's

50:38probably the best search platform, but it

50:40started off pretty average. And then over

50:42time, it's been built in public, and it's

50:44gotten better and better and say, Yeah, now

50:47, now, now it is what it is YouTube

50:49algorithm is a similar thing, right? And

50:50tick tock algorithm, all these things, they

50:53built in public. So it's amazing. This has

50:56been really, really useful. I think what

50:58this has taught me is that

51:00this is something I've been thinking for a

51:02while, which is like, the Tableau developer

51:05is becoming a really interesting role.

51:08Because on one side, you've got this, what

51:11I consider the exploratory sort of person,

51:14the person who sent out to go and find

51:16insight, then you've got the now analyst

51:19who just builds maybe ad hoc reports for

51:23their own purpose, so they can talk to

51:25people and tell them what's going on. Then

51:27you've got consumers who are consuming all

51:28of this content, and then you've got the

51:28who are consuming all of this content,

51:30wherever it's built, whether it's the

51:31analyst, he's built or the developer has

51:33built it. But I think in the past, we just

51:36used to say, I, you know, everyone who

51:38works in Tableau desktop can do all of

51:40these things. And actually, this highlights

51:43to me that now there are starting to become

51:45disciplines in Tableau, you know, where if

51:48you're good at, I'm just going to call a

51:51statistical analysis, data science, and you

51:54're familiar with the tools in that space,

51:57then, you know, this is the area to really

52:00focus on, invest your time and become

52:02almost an expert in this. And don't worry

52:04about all the other stuff like dashboarding

52:06, design and stuff, right? Just become good

52:09at this. And so other people will be good

52:11at dashboards, and other people will be

52:13good at, you know, writing calculations.

52:16And the bit that's missing is the

52:18collaboration capability. Like we all can't

52:21work on the same workbook at the same time.

52:23That's, that's the difficult thing, right?

52:25And when Tableau

52:26brings that capability, I think we'll start

52:29to see that sort of happen a bit more,

52:31where someone can come in, build the

52:33connection. And once the connection is

52:35built, the person who really understands

52:37the data or the analyst can put together,

52:39you know, a couple of quick charts, then a

52:42UX designer can come in and make that

52:44interactive and really nice. And everyone

52:46can collaborate at the same time. You know,

52:50I think you're right. I also think that the

52:53way that Tableau is seen, you know, is

52:56pretty much in silos, like, okay, you have

52:59people that have a role of data creation,

53:02data source creation, so the published data

53:04source, and then from there, explorers and

53:07creators can access data source. And they

53:10will focus on either doing exploration or

53:15making dashboards that will be used day to

53:18day.

53:18And it's my, to my experience, that's what

53:20I see most of the time, right? The data

53:22exploration part is less used, because

53:25people see Tableau as the dashboard slash

53:28report tool, where before I was doing it in

53:31my pivot table in Excel, no, I'm actually

53:34doing it in dashboard on Tableau.

53:38But again, yeah, it's very siloed, like,

53:40okay, you have dashboard creation, and then

53:43you have data expression, while the two

53:45could be linked together. And I do tend to

53:47see more and more people going in that way,

53:49like, I need to build a dashboard. But

53:52first, I want to do some exploration in my

53:53data.

53:56And like you said, yeah, good roles, roles

53:59are evolving. And I think it's getting

54:01better and better. And through the Sales

54:04force, there's more salesforce. Yeah, there

54:06's more to do in through Salesforce, it's

54:09getting even more and more to do because

54:11Tableau is really getting integrated into

54:14Salesforce. And yeah, I'm not even into the

54:18Salesforce world right now. But I'm pretty

54:20sure that you've heard of CRM analytics,

54:24which is also a big game changer.

54:26Okay, so that's it. I've actually decided

54:28to cut the video right here, because we

54:30went on to talk about a bunch of other

54:31things that have nothing to do with table

54:32extensions, and the video was already long

54:34enough. But actually, if you're interested

54:36in watching the rest of the video, it's

54:37about 15 minutes is something that I've

54:38uploaded to our members. So essentially, if

54:41you're a member of the channel, you can

54:42actually see some of the behind the stuff

54:44content that I create for the channel, and

54:46stuff I get up to around the YouTube and

54:48making content here. So I've uploaded it

54:50for members of the channel so they can

54:52check it out. But anyway, thanks for

54:54watching this rather different video.

54:55If you've got any comments or opinion or

54:57feedback about the video, let me know below

54:59. I'm also going to make a couple more

55:00videos about table extensions. This is by

55:02far not it at all. So there's so many

55:04different options. I spoke to so many

55:06different people about these. And I've also

55:08got a couple of use cases that Tableau sent

55:10through that I'm going to get stuck into

55:12next week once we start to actually

55:13understand how to use this. And once I'm a

55:15little better versed at using the

55:17capability. Thanks for watching, and I'll

55:19catch you in the next video.

55:20Transcribed by https://otter.ai

55:26[ Silence ]