Predictive model improvements

0:00Hey, it's Tim here and in 2020.4 Tableau

0:03have added the ability to now work with

0:05different

0:05predictive modeling functions inside of

0:08Tableau Desktop and the browser as well. To

0:11show you

0:11this I'm actually going to start off in

0:12Tableau Desktop. We're going to connect to

0:13Superstore

0:14Sales and we're going to build a very basic

0:16line chart. If you haven't seen my other

0:19video

0:19on extending the date access range that is

0:21a really useful video to watch before you

0:23watch this one because I will be doing that

0:25a little bit here and it's just good

0:27context

0:27for what's going on in this video so I won

0:29't be covering it so go watch that video and

0:32I'll see you back in a bit. I'm going to

0:34connect to Superstore Sales here and we're

0:36going to

0:36hop right in essentially. We're going to

0:38wait for this to load and we'll have our

0:40data set

0:40pretty much ready to go. Now the key thing

0:42I want to do here is just build out a

0:44simple

0:44line chart and I'm not going to try and do

0:47too much education of predictive models

0:49here.

0:50I'm just going to try and show you the

0:52predictive models and show you what's new.

0:54So the first

0:54thing I'm going to do is drag sales onto

0:56rows and then I'm going to drag order date

0:59onto

0:59columns. That by default brings in a

1:01discrete year if that makes sense. I'll

1:03make it a continuous

1:04month and that gives us this nice timeline

1:07we've got here. Now in 2020.3 they added

1:11the

1:11ability to do model quantile and model

1:14percentile. And essentially these are sort

1:17of methods

1:17of doing predictive analysis. They're sort

1:19of two functions as you would call it in

1:21Tableau

1:22terms. And so what we're going to do is we

1:23're actually just going to use one of them.

1:25We're

1:25going to use the model quantile one because

1:28that is actually quite a straightforward

1:30one

1:30to show on the screen and explain. The

1:32percentile one is more of a range based

1:35sort of predictive

1:36model. So with the quantile one what we can

1:39do is we can choose the median and then we

1:41can see how that actually behaves because

1:43the median is supposed to reflect

1:45essentially

1:46the middle of the data. So 50% of the data

1:48should fall above it and roughly 50% should

1:51fall below it according to the model. So

1:53let's go ahead and create a calculation and

1:55we're

1:56just going to bring this here into the

1:57middle and we have our calculation window

1:59ready to

2:00go. Now one of the things you can do is if

2:02you don't know how to write these functions

2:03you can just type in the name model in this

2:05case and you'll see model percentile and

2:08model

2:08quantile. We're going to use quantile here

2:11in this particular case and it's going to

2:13actually give us some hint as to how the

2:15model should be written if that makes sense

2:17. So

2:18what I'm going to do is I'm just going to

2:20write it as if I'm not putting any

2:21variables

2:22into this. I'm just using the standard

2:24function which in this case is a linear

2:26regression.

2:27So you can see this sort of nomenclature

2:30here. First thing it wants is the quantile.

2:32So if

2:32I just type in 0.5 for the median that's

2:35going to work quite nicely and then the

2:37target expression

2:38is the sum of sales in this case. In the

2:40other video you'll see that I suggested

2:42that we

2:43shouldn't be actually typing this out. I'm

2:45quite comfortable with Tableau so I know

2:46how

2:46to do this myself but if you want to make

2:48sure you don't get it wrong then what you

2:51want to do is drag them from the menu bar.

2:52In fact that's what I'm going to do for

2:54this

2:54next one because the predictor expression

2:56in this case is the month date and I need

2:58to make sure it's the correct one. So let

3:00me just click and drag here. I'm holding

3:02command

3:02on the Mac while I do that. On a Windows it

3:04's control and we pretty much have

3:06everything

3:07we need here. The last thing is obviously

3:09this date truncation is causing us a

3:11problem.

3:12You'll see here we get the error cannot

3:13makes aggregate and non aggregate arguments

3:15with

3:16this function. We've got sum of sales here

3:18which is an aggregate. The model is

3:21basically

3:21encapsulating all of this so the thing we

3:23actually need to fix here is the date trunc

3:25ation.

3:26So the way to sort of make this an

3:27aggregation is to throw an attribute at it

3:29essentially.

3:30So ATTR that's a good old trick. If you've

3:32used Tableau for a while you know that

3:34trick

3:34basically. If you need to aggregate text or

3:37something like that attribute is an

3:39absolute

3:39winner. In this case it makes a lot of

3:41sense because we're working with dates and

3:43the dates

3:43alike. I think of them as buckets so we

3:45need to actually just attribute it to the

3:47whole

3:48month not just the particular day if that

3:50makes sense. I mean I probably murdered

3:52that

3:52description so let's not go into that in

3:54this video maybe another video on that but

3:56long

3:56story short we need to make sure that it's

3:59all an aggregate. So here we have our model

4:01and I'll just call this example one and I'm

4:05actually going to call this linear

4:06regression.

4:07I'm just going to call this linear R

4:09because I know it is. I'll show you how I

4:12know that

4:13it is in a second so hit apply and what we

4:15'll do is we'll actually put this onto the

4:17canvas

4:17and I won't close the calculation window so

4:20let me just make this a little bit smaller

4:23and I'll bring this down here just off

4:24screen and we'll actually drag the example

4:27one right

4:27next to sales and this will actually draw

4:30another chart. You can see the two lines

4:33there

4:33and what we can do is we can actually dual

4:35access these so that they lie on top of

4:37each

4:37other. The last thing to do of course is to

4:39synchronize the axis so it's really really

4:41clear and now we have it. So the whole

4:44point of this linear regression and the

4:46quantile

4:46that I've chosen is that essentially this

4:48model predicts that half the values are

4:51going

4:51to fall above and below so it's the exact

4:54middle of the data set and it generally

4:56looks

4:56at the data set in a very simple way. Now

4:59the reason I know this is a linear

5:00regression

5:01is if I actually go to the Tableau

5:02documentation on this where they have

5:04advice on choosing

5:05a predictive model this is actually new in

5:072020.4 they actually have this page and the

5:10default function that came with model quant

5:13ile and model percentile was actually the

5:15linear

5:16regression so you don't actually need to

5:18specify it because it is the actual default

5:20. So if

5:21I go back to my calculation here and I just

5:23click put my mouse in here you'll see that

5:26if you look at this really really long

5:28expression here that the model has a

5:30different way of

5:31describing the model if you want to. So you

5:33can either do it the simple way which is on

5:35the left quantile target expression

5:38predictor expression or you can do model

5:41and then model

5:42quantile target expression and predictor

5:46expression. So model quantile target

5:49expression and predictor

5:51expression so this one on the right

5:53actually has the specificity of the model

5:55you want

5:56to run so we actually need to do that in

5:57this case to make things work a little bit

6:00better.

6:00Now in order to do that I'm actually going

6:02to split this out onto multiple lines it's

6:03a little bit easier to see. Okay so let's

6:05clean up this calculation I'm just going to

6:07go up to this 0.5 I'm actually going to go

6:10onto a separate line like so then I'm going

6:12to go to the sum and I'm going to open up a

6:14separate line then I'm going to go to the

6:16attribute I'm going to open up a separate

6:19line again and I'm going to go to the final

6:22I think that's pretty much it actually I

6:24don't need to do any more of that what I

6:26can do

6:26is I can put these brackets in new lines so

6:28that it's easy to know where things start

6:31and where they end. If you look at this

6:34date truncation actually what we can do is

6:36this

6:36should actually go back one here because

6:39this is all just one function and actually

6:41that's

6:42the last bracket here that actually we

6:44started over there. One thing you could do

6:46if you've

6:46got a bit of OCD is do that so that the

6:48brackets are in line then you can of course

6:51highlight

6:51everything and just tab them in one so that

6:54everything's sort of nice and neat. So this

6:57is great one of the new features in 2020.4

7:00is the ability to do block comments so the

7:02way you do that is you could do something

7:05like this forward slash star say something

7:07and then say more here and more and then

7:11you can close out on a new line with that

7:14and

7:15that actually does a block comment it's a

7:16really cool feature check out my video on

7:18that but here what I now want to do is add

7:21in the new functionality if I click in here

7:24you'll see that it gives me this ability to

7:26add a model so what I'm now going to do is

7:28I'm now going to specify a model. Okay in

7:31order to specify a model we just need to

7:32basically

7:32create a new line and I'm just going to do

7:35a double space there I'm just going to type

7:37in model and I'm going to do equals and I'm

7:40just going to do speech marks because that

7:42's

7:43typically what you do to enter text and I'm

7:45going to do a comma just to make sure that

7:47's

7:47working now the calculation still thinks it

7:50's got an error I think what I've done here

7:52is

7:52I've actually put the speech marks in the

7:58wrong place so if I do that there and then

8:04if I just type in linear in here let's just

8:07make sure this is correct sometimes I

8:09forget

8:09the notation if I get it wrong we can go to

8:12the documentation and search there we go I

8:14think I've got it right so what it didn't

8:16like is it didn't like the space that I had

8:18there before and then previously I didn't

8:20put the speech marks in the right place so

8:22it's very it's very fussy it's only going

8:24to sort of be happy once you've put

8:25everything

8:26in exactly the correct place now you can

8:28see here I've rewritten this in the new

8:31format

8:31I'm just going to move it here to the top

8:33right so we can see the chart more clearly

8:34here at the bottom and if I just hit apply

8:36you'll see that nothing changes because

8:39this

8:39is the same model we were using before it's

8:41actually the default model okay so now we

8:44can actually start playing with some of the

8:46new models in 2020.4 so the new one one of

8:48the new ones is a regular regular I can't

8:51even say this a regularized oh man this is

8:54a real tongue twister let's see how many

8:56times it takes for me to get this right

8:58regular

8:58regularized linear function there we go I

9:02got it so this one is actually quite easy

9:05to type thankfully you don't have to type

9:07that sort of mouthful you can just type in

9:08rl if I was to say in better terms this

9:11model will find a line that generally fits

9:14the trend

9:15it's not as fussy as the linear regression

9:17in terms of following the trend if that

9:19makes

9:19sense so let's just type in rl and hit

9:22apply and you'll see that it's ever so

9:25slight change

9:26there if I just do if I just copy this out

9:28here what I can actually do is I can be

9:31really

9:31smart I can actually comment out that line

9:34and then what I can do on a new line is

9:36copy

9:37this and what we can do is actually just

9:39comment in and then comment out pieces of

9:43the of the

9:43calculation so let's just paste this here

9:46and we can say that this is linear okay and

9:49now if I hit apply because this has been

9:51commented out it won't work so if I do it

9:53apply you'll

9:54see the line shifts ever so slightly it's

9:56the most subtle of changes and then if I

9:58cross

9:58that out and go down to a new line double

10:00forward that it's commented out you see the

10:03line changes back so it's probably not a

10:05good example here this probably will work

10:08better

10:08in a slightly different sort of time frame

10:11but essentially that model is is now

10:13available

10:14for you to try and then the last one we're

10:16going to try is a gaussian model so let's

10:19just go in here and hit enter go in again I

10:21'm going to use a double space I'm going

10:24to comment out the first one go forward

10:26slash forward slash and then comment out

10:29the last

10:29one dash dash and what I want to do here is

10:33the terminology for this I forget this so

10:36I'm just looking up on my second screen

10:39here is gp so gp is the actual expression

10:42so let's

10:42go ahead there and hit apply and you can

10:45see that this is a very different model it

10:47actually

10:48pretty much follows our trend in a very

10:51interesting way and so there we have it we

10:53have the two

10:54new models that have been added the regular

10:58ized I still can't say regularized linear

11:01model

11:01and the gaussian model that have been added

11:03alongside the linear model which was the

11:05previous

11:05default and you can of course use this for

11:07the model quantile and model percentile

11:10functions

11:10that are now available as predictive

11:13functions inside of the calculation window

11:15or tableau

11:16desktop or even in the browser so that's

11:17pretty much it that's the feature in a

11:19nutshell I

11:20highly encourage you to actually go to

11:22tableau's documentation I know I'm a little

11:24bit of a

11:25shill for tableau's documentation I'm

11:27probably a shill for tableau but there's

11:30great great

11:30content here in the documentation I can't

11:33emphasize enough how often people come to

11:35me for help and I just go to the

11:37documentation I feel really bad when I send

11:39people the document

11:41as if to suggest that they they hadn't

11:43thought to read it but it is an easy

11:45mistake to make

11:46you wouldn't expect a software company to

11:48go into this much detail in terms of

11:50describing

11:51features that you know are actually

11:53generally not exclusive to the piece of

11:55software these

11:56models are are not necessarily exclusive to

11:59tableau they can be found in any

12:00statistical

12:01tool pretty much but as you can see here

12:03tableau have done a nice job of really

12:05summarizing

12:06when you can use this model so linear

12:08regression is the default as I said to you

12:10earlier on

12:10use when you have only one predictor and

12:12the predictor has a linear relationship

12:13with your

12:14target use the regularized linear

12:16regression I mean they've got to come up

12:18with a better

12:19name for that come on use when you have

12:21multiple predictors especially when the

12:23predictors

12:23have linear relationship to the target

12:25metric and those predictors are likely

12:27affected by

12:28similar underlying relationships or trend I

12:30mean I just zoned out after the first

12:32sentence

12:32there but if you're into this stuff and you

12:34're being asked are you being pressed to

12:37start

12:37to use these functions or use these

12:39capabilities in some of the way you do your

12:41work then it's

12:42nice to be able to have them available to

12:44you and of course you can then start to you

12:46know put them out there and see how people

12:48challenge the models or choose which model

12:50is the best one and so on and so forth but

12:52it's just great to have these features

12:54there

12:54and you can see that you know it's

12:56reasonably easy to add them to tableau to

12:58be brutally

12:59honest it's not a ton of work and not like

13:01R or something else where you have to do a

13:03lot of data framing and sort of description

13:06of what you're about to do and a tableau

13:08calculation

13:09is super super simple so anyway I'm r

13:11ambling on now let's end this video and if

13:13you've

13:13enjoyed the video you know what to do if

13:15not drop a comment below let me know what

13:18you'd

13:18like to see instead or send it to someone

13:20you don't like they might appreciate it

13:21more

13:22thank you I'll catch you in the next one.