Cyprus TUG 19 May 2026
Tableau Mastery Offline: Calculations, Extensions & Affiliate Analytics
42 min 4,509 words transcribed
Summary
Recorded at the Cyprus Tableau User Group's in-person meetup in Limassol, "Tableau Mastery Offline: Calculations, Extensions & Affiliate Analytics", and published on the Cyprus Tableau User Group YouTube channel. Vasily Lavrov, Senior Business Intelligence Analyst at Exness, gives a workshop-style deep dive into Tableau table calculations for people who already use them regularly.
The central idea is that a table calculation's result depends on two things: the formula and how it is computed. Vasily demonstrates Compute Using with top-N ranking and difference-from calculations, compares each with the equivalent SQL window function, and explains the rule the interface hides: dimensions you leave unticked become the partition, and dimensions you tick set the direction of the calculation. TOTAL is compared with window functions and with EXCLUDE and FIXED LOD expressions, including how dimension filters affect each in the order of operations. PREVIOUS_VALUE is shown filling gaps in a time series and, as a self-referring calculation, building a list of every value selected in a filter, kept intact by filtering on LAST() = 0. The talk ends with nested table calculations, which can each have their own direction when kept in separate calculated fields.
In the Q&A, the audience asks about the performance of table calculations against LOD expressions, filling large gaps, when to move logic out of Tableau into the data layer, and whether LLMs are useful for writing calculated fields.
Key points
- A table calculation's result depends on the formula and on Compute Using.
- Unticked dimensions partition the data; ticked dimensions set the direction of the calculation.
- Table calculations map closely onto SQL window functions: PARTITION BY and ORDER BY.
- TOTAL behaves much like an EXCLUDE LOD, recalculating the aggregate across the partition.
- Dimension filters affect TOTAL and EXCLUDE but not FIXED; context filters to work around that add complexity.
- PREVIOUS_VALUE refers to the calculation's own previous result, so it can carry values forward and build up lists.
- Filter on LAST() = 0 to keep one mark without breaking a table calculation.
- Keep nested table calculations in separate fields so each can have its own direction.
- Move long-lived or heavy logic upstream into the data; use Tableau for quick prototypes.
Jump to a chapter (19)
- 0:00 Welcome and what the workshop covers
- 1:35 Formula plus Compute Using
- 2:40 Top N with RANK
- 4:37 Table down versus pane down
- 6:10 The comparison with SQL window functions
- 7:00 Partitioning and addressing
- 7:41 Difference from previous quarter and previous year
- 11:10 TOTAL
- 15:17 TOTAL versus EXCLUDE and FIXED
- 16:30 Order of operations and filters
- 19:40 PREVIOUS_VALUE: filling gaps
- 24:34 Listing every selected filter value
- 27:37 Keeping one mark with LAST() = 0
- 29:07 Nested table calculations
- 31:55 Thanks and a prize
- 33:58 Q&A: table calculations or LOD for performance?
- 36:10 Q&A: large gaps and when to leave them visible
- 38:28 Q&A: when to move logic into the data layer
- 40:37 Q&A: LLMs for calculated fields
Transcript
0:01 Vasily Lavrov: I, I should move my crown a little bit. Uh, hello everyone. Uh, I'm really glad to see you all here, and, uh, I'm glad to open our today's, uh, Tableau User Group.
0:14 Uh, my topic will be quite technical, I suppose. Target audience is, uh, uh, those ones who work with Tableau quite regularly and, uh, familiar with table calculations.
0:30 Uh, it will be more in the format of a workshop, so if you have any questions and, or want me to clarify something, feel free to ask.
0:40 If it will be quick, I will answer immediately. If not, I will postpone it till, uh, the Q&A session or break or networking part.
0:49 Uh, I will show a lot of examples, uh, in Tableau Desktop.
0:54 So if you wish, you, uh, may download the workbook, uh, right now, but I suggest just to listen and watch and, and then, uh, if you miss something or want to go deeper into some cases, you may download it from, uh, Tableau Public, um, and check.
1:16 Uh, first part will be, uh, about Compute Using. So it's how to set up table calculations on your view.
1:35 Compare table, um, uh, to make it easier to understand. Uh.
1:44 Then speak about how to manually set, uh, direction of calculation.
1:55 Uh, and I will start with one outcome.
1:59 Uh, presentation just started, but I already want you to, uh, see this idea and to keep it with you when you go home.
2:07 Table calculation, its result is defined by two parts. First, calculate.
2:16 And Compute Using. That's how you set order, direction of calculation on your view.
2:24 So, for example, one calculated field may return different results, uh, if you compute it in different ways.
2:34 Uh, let's go, uh, into Tableau Desktop, and I will show it, um, here.
2:40 First example is, uh, quite basic.
2:44 I think it's one of the popular usage of, uh, table calculations is, uh, calculating top N, uh, elements. So for example, we have list of subcategories. I want to order them by sum of sales and, uh, show only top three.
3:03 Who did something similar in your experience before?
3:09 Yeah, good.
3:11 Uh, here, um,
3:16 I u- use two, uh, calculated fields.
3:21 First is rank, which is enumerates my subcategories.
3:26 Uh, table calculations, uh, works the following way.
3:30 First, uh, Tableau calculates, uh, like simple aggregations on each mark, and then rank. It is a table calculation. It is another layer, another, uh, aggregation above pre-calculated arguments.
3:49 Like rank function in SQL, for example. It works pretty the same.
3:54 Uh, and the next, uh, field I use is, uh…
3:59 Oops, no.
4:04 Uh, filter by top three.
4:09 Um, I use rank from the previous step and l- leave only three values.
4:23 Compute Using. When you put table, you can set, uh, a computation. Table down, pane down.
4:30 So here I use table down.
4:33 It means that Tableau goes into one partition across all, uh, marks on my view.
4:42 Let's see a little bit more complicated example when we have two dimensions. And now I want to show top three subcategories within each category.
5:02 So without filter, only on the first category.
5:06 I use absolutely the same rank, rank function, but I go, uh, into compute, Compute Using, and let's look what we'll have, uh, if I show, uh, it, uh, table down.
5:22 Uh, you see rank enumerates all marks in, in partition.
5:28 If I use pane down, for example, Tableau calculates ranks i- independently inside each pane, inside, uh, separate categories.
5:40 Uh, let's have a look, um, a comparison with, uh, queries.
5:48 So for example, when I, uh, comp– make Compute Using pane down or…
6:01 When I manually click rank We can, we can, um, SQL like this. Select category, subcategory, category, sum of sales, like, uh, simple aggregation and rank. It's a window function which is calculated, uh, above pre-calculated total sales and partitioned by category.
6:22 But quite similar, uh, in Tableau.
6:29 Uh, let me show one more time. When I go to edit table calculation, uh, you will have, uh, basic built-in Compute Using like table down, paid down, but also have specific dimension.
6:43 When I started to work with Tableau and didn't know how it work, I was just clicking it randomly, unsuccessfully, trying to understand how it works.
6:51 Who did, uh, the same? Who did?
6:55 Yeah, yeah.
6:56 Um, so now I will share the secret how it works. It's not visible in the interface.
7:03 Uh, actually, when you s-
7:12 … fields, you split them into two groups. Select, they goes into partitioning, meaning they split your marks into separate partitions isolated from each other.
7:24 And, uh, others which you pick, uh, which you select, they go to addressing, uh, meaning that Tableau will calculate along these fields.
7:37 Let's go to the next example.
7:41 Uh, difference from. Imagine, uh, I have the following table with sales for each quarter,
7:51 and I want to calculate difference, uh, from
7:57 previous quarter, like Q4 minus Q3, plus something, and difference from the same quarter but one year ago. So Q4 24 minus, uh, Q4 23.
8:12 I use one field for both calculations. Let me show the formula.
8:20 Um,
8:24 so it's sum sales, which is calculated on each mark, minus lookup minus one. Lookup is a function similar to lead or lag, uh, in like PostgreSQL, which can, uh, go like one or several steps, uh, behind or, uh, further in your partition.
8:46 So one formula, but I use different Compute Using.
8:54 Difference from previous quarter, it's table across or in terms of Specific Dimensions, uh, I select both year and quarter, meaning that I don't have, uh, uh, several partitions. I have only one partition, and I go through it and calculate lookup minus one from previous mark.
9:19 Uh, and for another, uh, it's more tricky.
9:26 I split my, uh, view by quarters into isolated partitions and, uh, go along the year.
9:38 So it's like we split it into quarters and make one step back.
9:47 Uh, and again, uh, let's compare it with, um, very simple.
9:54 Uh, when it's difference from previous quarter, we have only one partition and go order by year and order by quarter. So no partition by.
10:06 Uh, when we want difference from previous year, so partition by quarter, order by year.
10:12 Uh, you may check how it works in Tableau, may check how it l- looks in your database.
10:20 So probably the results will be same.
10:25 Okay.
10:28 Now I think I'll go to the next part. Uh, but again, before– uh, one more outcome.
10:36 So how to remember what these, uh, checkboxes, uh, do.
10:42 If you do not select a field, uh, a dimension, it goes to partitioning and divides data into isolated slides.
10:50 If you select dimension, it determines the direction of, uh, calculation, so which, uh, field you go first, which next.
11:02 Any questions maybe?
11:04 I, uh…
11:06 Very intuitive.
11:08 Now or before?
11:11 Yeah. Uh, yeah, yeah, I understand.
11:13 Yes. Actually, I think I found out it, um, not from Tableau knowledge base, but from some articles of enthusiasts, and it really helped.
11:25 Uh, okay.
11:26 Next, total.
11:28 I really like this function. I think it's the most popular table calculation. I use it, uh, quite regularly.
11:36 Uh, and, um, the most popular example, if calculating
11:48 mark, and for denominator, you want to calculate total across, uh, several marks.
11:52 Uh, the interesting part of total is that it works, uh, differently from other table calculations.
12:00 Uh, I will compare total with, uh, window functions like window sum, window average, uh, and I will compare it with LOD, level of details calculation.
12:12 My, uh, understanding of total, the message, that total is works more like an exclude rather than a table calculation, but total was available, uh, much before LOD appeared. So I'm from the era before LOD, so I really like total and do not like LOD.
12:34 Uh, who, who, who, who is a LOD person? Who is the, uh, table calculation? Who prefers LOD?
12:43 I see.
12:45 Unfortunately, some of my colleagues.
12:48 I, I should talk with you strictly.
12:50 Um, okay, okay. Uh, just kidding.
12:57 So, uh, for total, imagine the following table. We have, uh,
13:06 some categories and want to calculate how many unique customers made at least one order from this category, and then calculate percent of, of, of these customers from total client base.
13:19 The question, how to calculate total client base?
13:22 Uh, uh, window function.
13:35 Just aggregate, aggregations, sorry, um, on each mark, and then make like second calculation above these marks. So i-if I calculate window sum of count of customers, Tableau just, uh, calculate with these numbers, which is not correct because, uh, we may have, uh, intersections.
13:55 Uh, one customer may have, uh, orders from different categories.
14:00 The correct answer will be using total.
14:05 Hmm.
14:06 Let's look at total function.
14:11 Total itself, uh, doesn't have specific aggregation like window sum, window average. It's just total. It…
14:20 Actually, it breaks your detailization and calculate, uh, initial aggregation from its argument like we didn't have, um, split by dimension.
14:34 So if I remove all d- all dimensions, total will be 800.
14:41 If I return it back, total will be the same.
14:45 Uh,
14:48 let's look at Compute Using. Yes, the table down, meaning we have only one partition, and we calculate total inside this partition.
14:56 So yeah, total ignores your level of details and calculate, um, row-level data inside partition, which sounds similar to what exclude do, does.
15:11 Uh, let's now compare total versus, uh, level of details calculation.
15:17 Uh, so one total calculated using table calculation total.
15:24 Uh, one using exclude. The instruction for Tableau is exclude category from the view and calculate count of customers as if we didn't have any split.
15:38 So it also return 800.
15:42 And, and the third one, ex…
15:50 instruction, uh, for Tableau for fixed, it's like fixed ignores absolutely everything, uh, all, all dimensions on your, uh, view and, uh, o- only calculate, uh, a- aggregate data using only specified dimensions. In my case, I want to calculate the whole client base, so I don't need to fix it by, um, any dimensions.
16:15 So we have three options how to calculate total.
16:20 Now, it returns the same numbers. Why?
16:24 Because of, uh, we don't have additional, uh, limitations. Let's look at order of operations.
16:31 Who knows, uh, similar, uh, pictures?
16:36 Yeah, yeah. Uh, I suggest, uh,
17:01 before dimension filters. The dimension filters do not, mm, affect it, and exclude in the middle.
17:09 So if we start to adding additional conditions, these results may differ.
17:15 For example, let's filter only one category.
17:20 You see that, mm, these totals remains the same. Oh, not remains the same, but, uh, recalculated inside category.
17:29 But fixed still show the same result. Why?
17:33 Because dimension filters do not affect fixed LOD.
17:37 Sometimes it's, uh, really, uh, required behavior.
17:43 But sometimes I see that people use fixed, apply filters, and then make these filters, uh, make, make them context filters.
17:55 So context filters, uh, affect fixed LOD.
17:59 I do not suggest such approach because, uh, you apply unnecessary complexity to your, uh, workbook. So s- since you make such approach, you need to put all your dimension filters in context if you want them to affect LOD calculations.
18:25 Uh, I think that's all I want to say about total.
18:29 Of course, there are a lot of nice examples more, but I have limited by time. Any questions, additions?
18:43 So I go to the, uh…
18:46 Ah, first of all, yeah, I already told about my preferences.
18:52 I prefer total.
18:53 Sometimes I use exclude, and I try not to use fixed at all unless it's, uh, really needed by the design of my workbook. For example, when I want, uh, fixed not be affected by dimension field.
19:09 Uh, and outcome for this part, uh, what remember about total is that to-
19:23 it ignores and aggregate its argument across all row-level data within the partition, uh, you set for it.
19:32 Roughly speaking, it be- behaves similar as exclude LOD.
19:37 But of course, uh, in different conditions, uh, based on order of operation, you may get different results.
19:49 Okay, next part, previous value.
19:53 This is… Unlike total, this is super
20:02 rare, uh, function. Anyone who used it different?
20:08 Once. Once.
20:10 Uh, I remember I used it only once in Exness during last four years, and once I consult, consult my colleague about how to use it.
20:20 Uh, the best way to un-
20:30 I think it's to come as one, uh, previous value, but we should understand previous value of what.
20:37 And, uh, also I will show you one, um, example illustrating that previous value works with text well, and this is really funny. Sometimes I, I use it.
20:47 Um, maybe it will be helpful for you too.
20:53 Okay, let's go back to Tableau.
20:56 Uh, the following example.
20:59 Uh, I have a time series, uh, of a price, but it has gaps.
21:07 Maybe some data is missing, maybe markets were closed and we don't have data.
21:12 So I want this chart to look like this, so to fill gaps with last available value.
21:22 Uh, I can do it using previous value.
21:27 The formula is the following. If now, uh, it works like Coalesce in PostgreSQL, I think. Uh, it…
21:35 We take first argument, but if it's null, if it's empty, we take the second argument.
21:44 So we go, uh, across partition, across, uh, all, um, dates.
21:52 Take data.
21:54 When we miss data, instead of price, we take its previous value.
22:01 The question, why we can't do it with lookup minus one?
22:07 Let's check it, uh, in the table, what will be the difference.
22:13 Look at these two, uh, fields. First price, like simple aggregation, and then lookup minus one. Table calculation which takes value from, uh, previous, uh, step in partition.
22:30 But it takes not previous value of itself of a calculation. It takes previous value of argument.
22:37 So you see it's just shifted one step ahead.
22:42 Uh, and if we repeat the same calculation using lookup,
22:52 so we'll take price when we have it. When it's empty, we take the previous argument, lookup price minus one.
23:01 But if we have several gaps in a row, we still continue take the previous result of an argument. So it can take only one, uh, previous step, not, not, not, uh, the whole history.
23:17 And here is the main difference how lookup works and previous value works.
23:21 Uh, because previous value is like self-referring function. Previous value take previous value of the whole calculation of itself, not the previous value of, um, argument.
23:39 And main idea to remember about previous value is that it's self-referral, a curs- kind of a recursive function unless as a table calculation. Meaning that previous value refers to the previous result of calculation, not the previous result of, uh, an expression argument, but the whole calculation.
24:02 Uh, I think that's all you should remember. In case you realize sometimes that you need something iterative, like to refer data from the previous, uh, calculation step, uh, you should remember about previous value and then you can Google and maybe f- find your, um, solution for your case
24:25 And now, uh, funny practical example.
24:31 Imagine we have some dashboard, we have filter where we select some, uh, values, and I want my users to see the whole list of chosen values.
24:46 Uh, actually, we have a built-in function. For example, we may ask Tableau in caption or in title to show, uh, the chosen filter, but it will be work, uh, as follows. It will show one to three and some more.
25:04 And if we select all, it just say all.
25:09 So sometimes it helps, but if you want barely to show all chosen results, we m- can make it custom.
25:20 Let's go to this, uh, s- worksheet.
25:24 I will break it and quickly build it from scratch.
25:38 So first of all, we have a list of all subcategories, and we have a filter which applied to multiple worksheet.
25:44 Uh, now we add the following calculation.
25:55 We ask Tableau to take– on each step, take previous value which you accumulated before, plus add subcategory.
26:07 What this part do?
26:09 It work its default value. When we are in the first mark of our partition, we'll take this empty set.
26:19 We can write something here, and you see it will be
26:26 as a default.
26:28 In my case, I don't need it.
26:30 So,
26:32 moving from start to the end of my partition, I accumulate the list of, uh, all subcategories, and at the end, at the last point, I have the whole list.
26:41 Now, I want to show only this mark.
26:45 If I just filter it…
26:49 Oh, sorry.
26:52 on it.
26:54 The calculation will break.
26:56 Why?
26:57 Who knows.
27:00 Because of order of operation.
27:03 If we apply dimension filter, it affects table c- table calculation, so we need to filter it somehow after table calculation.
27:14 So the, the idea, Tableau calculate, uh, the whole result, and then we filter.
27:21 Uh,
27:25 was this.
27:29 Uh-huh.
27:30 Uh, we may use the following trick.
27:34 There is a table calculation last which return us, uh, uh, it's enumerate all, uh, points, all marks on the list.
27:55 So zero is the last mark in partition, and then it's increased. We go back.
28:02 So if we apply last equals zero,
28:07 it will be true only for the last mark in partition.
28:11 And now, if we ask Tableau to filter, uh, to filter by last equals zero,
28:21 we'll get only one last mark, but we'll keep our table calculation because table calculation filters right after table calculations and do not affect, uh, them.
28:38 So that's the way how we may generate a list of text values.
28:45 I think sounds quite noisy. Thank you, Andre.
28:49 Um, the last part, and you are free.
28:55 Um, last but not least, uh, nested table calculation. It's really, uh, quick and easy, uh, but sometimes it is missed because, uh, again, not very visible from interface of Tableau desk that we have such opportunity.
29:18 Uh, the idea is this, uh, is table calculation called nested.
29:22 And for these nested calculations, we may use different Compute Using.
29:28 For example, we might first calculate percent of total, uh, for each year, meaning table down across regions. And then we may calculate percent difference above these percents going table across.
29:47 Uh, the idea here is to keep, uh, these nested table calculations in different, um, calculated fields.
29:57 So if we put it into one formula, w- we, we can't, uh, set, uh, different directions.
30:04 Uh, let me show.
30:07 Uh, the first one, percent of total, it's just sum of sales divided by total of sum of sales.
30:16 And difference to previous,
30:19 it's like percent of total calculated for each cell minus lookup minus one.
30:27 So we take this minus this and get this.
30:32 And now the magic is in edit table calculation.
30:37 So when we go to this menu, Tableau identifies that we have nested calculations inside, and appears the drop-down menu where for percent of total, I may ask Tableau to calculate the table down.
30:50 And then for difference to previous, I may ask Tableau to calculate the table across.
31:00 That's all.
31:01 Who used nested table calculations before?
31:13 Nice. Now, I hope there will be more loyal to nested table calculations users.
31:18 Here I have another example with four nested calculations, but I will not show it. You may download the workbook and the reverse engineering how it works, how it works.
31:30 Uh, as usual, I finish with one idea. Keep nested table calculations in separate fields.
31:39 In this case, you will be able to set independent directions, independent compute.
31:46 That's all. Thank you for your attention.
31:55 Meetup host: Thank you very much, Vasya, for your awesome presentation. It was very useful. It's real magic with table calculations.
32:03 Vasily Lavrov: Magic and mystery.
32:05 Meetup host: Yeah.
32:06 Yeah. And now we have a few minutes– Ah, sorry.
32:09 Uh, we have a very beautiful gift for you. It's, uh…
32:30 Vasily Lavrov: Thank you.
32:32 Thank you.
32:33 Meetup host: And few minutes for, um-
32:35 Vasily Lavrov: Or sh- or sharing the opinions instead of questions.
32:37 Meetup host: Yeah.
32:37 Vasily Lavrov: It's also welcomed.
32:38 Meetup host: Yeah. Yeah.
32:39 And to make things a little more interesting tonight, uh, we have some exclusive, uh, swag for you. It's, uh, this cap from Tableau, the same like I have.
32:56 Yeah.
32:57 And, um, I actually receive, uh, this exact hat, hat, uh, in 2023 when I was a speaker at the very first meetup here, um, at Porta Bella Rooftop.
33:13 So, um, but the real reward, uh, was sharing knowledge with the community.
33:20 Um, and we are always looking for new speakers, um, for our future offline and virtual meetups.
33:28 Uh, so if you ever want to be a speaker at our meetups, uh, just let me know.
33:35 And now, uh, let's see who will take this, uh, hat home t-tonight.
33:44 Eduard.
33:46 Audience member: I don't need a cap. I just want
33:50 This is not my size.
33:52 Vasily Lavrov: You already have several caps, I suppose.
33:58 Audience member: No, I really, I would like to ask a question.
33:58 Vasily Lavrov: Duda, will you share a mic?
34:00 Meetup host: Yes.
34:01 Vasily Lavrov: Or, or let him try, uh, shout.
34:04 Audience member: Thanks.
34:05 Um, I've got one question. You said you prefer to have, uh, to work with the table calculations, uh, against LOD.
34:13 So have you ever tried to record– to, uh, make this, uh, performance recording to really investigate, uh, what is the difference between, uh, LOD or table calculations i-in terms of, uh, performance?
34:28 Vasily Lavrov: Mm-hmm. Mm-hmm. Mm-hmm. Uh, thank you.
34:31 Um, I think I didn't r-record, uh, and, uh, compare, um, in case of like share some insight about it. But I had some cases when, uh, removing LOD made, uh, co- uh, like overall performance, uh, better.
34:50 Uh, but I may say that, mm, usually, um, if you don't have very gra-granular fields with, with– Like if you don't show a lot of marks on your view, uh, table calculation feels like working better than LOD.
35:10 Uh, but there are, uh, cases when table calculations may works worse if you have a lots of marks on your view, because in this case, e-even if you then filter them by table…
35:32 Because this case have, uh, calculation on each mark, uh, and only then displayed.
35:38 So yeah, I, I don't have numbers to say for sure, but in my like, uh, feeling, usually table calculations, uh, works better.
35:49 But most-mostly due to some limitations when u-users start apply a lot of context filters or using sets. Uh-huh.
35:58 I have spoken.
36:00 Meetup host: Thank you very much.
36:01 Vasily Lavrov: Thank you.
36:10 Audience member: Hello. My name Katerina. I have a question about gap.
36:14 Uh, I saw you, your formula, um, with recursive.
36:19 Vasily Lavrov: Mm-hmm.
36:19 Audience member: If we have huge gap, how it will work if we have unexpected, um, missing numbers?
36:28 Vasily Lavrov: Uh, y-y-y-you mean h- uh, ah, you mean, uh, huge in terms of, uh, on the axis? So like 1,001.
36:39 Yeah, in this case it will, uh, like just take the previous result.
36:44 But I think, uh, you may somehow, uh, smooth this by, uh, so not taking just previous value, but maybe combine it with, um,
36:59 with, um, s- s- somehow, um, make a coefficient to make it…
37:19 Uh, no, but you create partitions the same but addressing reverse side. So it will give us value but, but from the back and then combine these two.
37:29 It's my idea. I didn't check it, but I think you… It's this you may like refer to previous and previous like meaning next.
37:41 Audience member: Okay. Thank you.
37:43 Meetup host: Thank you.
37:44 Vasily Lavrov: Went again.
37:45 Meetup host: Who else have questions?
37:47 Who wants a hat?
37:52 Audience member: Just a small comment on that previous… Yeah. Sorry, just a small comment on that previous, uh, value thing that, um, I think we might need to, uh, use it thoughtfully because sometimes, uh, maybe it's more important to s- really just see the gaps rather than just fill the… Just to, uh, make graph look nicer.
38:14 Vasily Lavrov: Yeah. Yeah, sure. In this case, it's just like, um, an example to show the functionality.
38:22 Meetup host: Thank you.
38:28 Audience member: Yeah. Thank you for brilliant presentation actually. Uh, the question is, um, some kind of, uh, out of scope a little bit. Uh, when you understand that, uh, you have to stop writing a code inside calculated field-
38:44 Vasily Lavrov: Mm-hmm
38:45 Audience member: …
38:45 and make the calculation on the level of a data mart instead.
38:50 Vasily Lavrov: Mm.
38:51 Audience member: For, for me, it's kind of pain, uh, yeah, in some places. Uh, so yeah, I understand that I make calculated field when it's kinda 50 r- rows of code, it's better to move it to, uh, like SQL or, or Python or something-
39:06 Vasily Lavrov: Mm-hmm. Mm.
39:06 Audience member: … or Dako Airflow or something.
39:08 Vasily Lavrov: Yeah. Yeah.
39:10 Uh, thank you. I may say that, for example, if I would like to make this example with gaps, I would prefer to fill the gaps not in Tableau, but on data level.
39:24 Uh, in general, speaking about any cases, uh, there are two ways I think. When I make something…
39:35 Uh, so the question when I decide, uh, how I decide to split, yeah, the guideline. Uh, if it's something, uh, quick MVP, when I need something quickly, I prefer to take data immediately and then start, uh, working in Tableau because it's quicker. But if I understand that I make a solution for a long-term usage, in that case, I start move as much calculations from Tableau to the data.
40:04 Ideally, in Exness, uh, we try to have like one source,
40:19 table and data source. One that needs more, uh, it's just the aggregations, like sum average or sum divided by sum because, uh, you can calculate, uh, non-additive, um, on a data level. Mm-hmm.
40:31 Meetup host: Thank you. Anyone else?
40:37 Audience member: And another one, uh, do you actually rely on, uh, LLM like Copilot or something, uh, to create a calculated field based on, uh, example, uh, for my site, uh, sometimes I have legacy-
40:55 Vasily Lavrov: Mm-hmm
40:55 Audience member: … kind, uh, disaster legacy. Yeah. And when I take like SQL code and say, make a calculated field from it, it do something, it makes something really complicated and it works.
41:08 Vasily Lavrov: Mm-hmm.
41:08 Audience member: Yeah. But, uh, and sometimes I'm, uh, stop myself from, uh, doing that, uh, understanding that it's kind of black box. Do you use it?
41:18 Vasily Lavrov: Uh, actually no. For Tableau, not yet. Um, it's… For me, it's, uh, quicker to write it by mas- myself. In our team, uh, we try to use, um, uh, AI to write SQL, but again, as I know, it's like can generate something and then again human check it, review, and then merge i- if it's okay.
41:44 But with Tableau, uh, I personally didn't try. Uh, I, uh, while I was preparing to this presentation, I asked the Tableau some cases about Tableau, uh, and Gemini provided me some examples. I checked them. One was correct, one was incorrect. So again, it's like humans should touch, uh, should check.