Обсуждение: Summing & Grouping in a Hierarchical Structure

> From: pgsql-sql-owner@postgresql.org [mailto:pgsql-sql-owner@postgresql.org] On Behalf Of Don Parris
> Sent: Thursday, February 14, 2013 8:58 PM
> To: pgsql-sql@postgresql.org
> Subject: [SQL] Summing & Grouping in a Hierarchical Structure
>

> Hi all,
> I posted to this list some time ago about working with a hierarchical category structure.   I had great difficulty with my problem and gave up for a time. 
> I recently returned to it and resolved a big part of it.  I have one step left to go, but at least I have solved this part.
>
> Here is the original thread (or one of them):
> http://www.postgresql.org/message-id/CAJ-7yonw4_qDCp-ZNYwEkR2jdLKeL8nfGc+-TLLSW=Rmo1VkbA@mail.gmail.com
>
>
> Here is my recent blog post about how I managed to show my expenses summed and grouped by a mid-level category:
> http://dcparris.net/2013/02/13/hierarchical-categories-rdbms/
>
>
> Specifically, I wanted to sum and group expenses according to categories, not just at the bottom tier, but at higher tiers, so as to show more summarized information. 
> A CEO primarily wants to know the sum total for all the business units, yet have the ability to drill down to more detailed levels if something is unusually high or low. 
> In my case, I could see the details, but not the summary.  Well now I can summarize by what I refer to as the 2nd-level categories.
> Anyway, I hope this helps someone, as I have come to appreciate - and I mean really appreciate - the challenge of working with hierarchical structures in a 2-dimensional RDBMS. 
> If anyone sees something I should explain better or in more depth, please let me know.
>
> Regards,
> Don
> --
> D.C. Parris, FMP, Linux+, ESL Certificate
> Minister, Security/FM Coordinator, Free Software Advocate
> http://dcparris.net/
> GPG Key ID: F5E179BE

My two cents would be to actually use a  different tool for the job of presenting this data.  I'd have used a pivot table in Microsoft Excel.  Not sure what your environment or requirements are but pivot tables are widely used in business, easy to share, can be formatted, and give the user the ability to drill down and navigate to the data they want to see.
I'd set up a query to pull the raw data you need with all of the categories and associated data you need.  Then bring that data to Excel to present and summarize it.

Mike

Hi,

Have you considered maybe ltree datatype?

http://www.postgresql.org/docs/9.1/static/ltree.html

I think it solves a lot of problems in topic....

Kind regards,

Misa

On Friday, February 15, 2013, Don Parris wrote:

Hi all,

I posted to this list some time ago about working with a hierarchical category structure.   I had great difficulty with my problem and gave up for a time.  I recently returned to it and resolved a big part of it.  I have one step left to go, but at least I have solved this part.

Here is the original thread (or one of them):
http://www.postgresql.org/message-id/CAJ-7yonw4_qDCp-ZNYwEkR2jdLKeL8nfGc+-TLLSW=Rmo1VkbA@mail.gmail.com

Here is my recent blog post about how I managed to show my expenses summed and grouped by a mid-level category:
http://dcparris.net/2013/02/13/hierarchical-categories-rdbms/


Specifically, I wanted to sum and group expenses according to categories, not just at the bottom tier, but at higher tiers, so as to show more summarized information.  A CEO primarily wants to know the sum total for all the business units, yet have the ability to drill down to more detailed levels if something is unusually high or low.  In my case, I could see the details, but not the summary.  Well now I can summarize by what I refer to as the 2nd-level categories.

Anyway, I hope this helps someone, as I have come to appreciate - and I mean really appreciate - the challenge of working with hierarchical structures in a 2-dimensional RDBMS.  If anyone sees something I should explain better or in more depth, please let me know.

Regards,

Don

--
D.C. Parris, FMP, Linux+, ESL Certificate
Minister, Security/FM Coordinator, Free Software Advocate

http://dcparris.net/

GPG Key ID: F5E179BE

>  That said, now that I have finally gotten the chance to try ltree, I think I like it a lot.

Hello Don,

Yes, after looking at ltree --which I had not done before-- I have to agree with Misa that it looks like the right solution for your problem.  That is not to say that "brute force" SQL couldn't provide a workable arrangement; but ltree looks very flexible, especially as it allows you to assign cost values to non-terminal nodes.  If it were me, though, I'd still make use of VIEWs to report results of the workhorse queries:  staring at a list of items like "Transportation.Bicycle.Gear.Chain_ring" sounds like headache.  That's a matter of taste, of course.

Bryan

Hi Don,

Yes, its better to use it in category table...

Bryan, how many levels there will be - we dont know...

With one table - and ltree is solved all cases...

To add new subcategory user just picks the parent category... So it is easy to add chain ring to gear... As category...

In another transaction table is category_id, amount...

Don already posted query for sum... In these case just category and transaction table should be joined  sum amount, group by functions on lpath....(depending what is the goal...)

Kind Regards,

Misa

On Saturday, February 23, 2013, Bryan L Nuse wrote:

This works fine:
test_ltree=> SELECT path, trans_amt FROM testcat;
                  path                   | trans_amt
-----------------------------------------+-----------
 TOP.Transportation.Auto.Fuel            |     50.00
 TOP.Transportation.Auto.Maintenance     |     30.00
 TOP.Transportation.Auto.Fuel            |     25.00
 TOP.Transportation.Bicycle.Gear         |     40.00
 TOP.Transportation.Bicycle.Gear         |     20.00
 TOP.Transportation.Fares.Bus            |     10.00
 TOP.Transportation.Fares.Train          |      5.00
 TOP.Groceries.Food.Beverages            |     30.00
 TOP.Groceries.Food.Fruit_Veggies        |     40.00
 TOP.Groceries.Food.Meat_Fish            |     80.00
 TOP.Groceries.Food.Grains_Cereals       |     30.00
 TOP.Groceries.Beverages.Alcohol.Beer    |     25.00
 TOP.Groceries.Beverages.Alcohol.Spirits |     10.00
 TOP.Groceries.Beverages.Alcohol.Wine    |     50.00
 TOP.Groceries.Beverages.Juice           |     45.00
 TOP.Groceries.Beverages.Other           |     15.00
(16 rows)

So if I want to see:

TOP.Groceries        | 240.00

TOP.Transportation | 180.00

Hello Don,

Perhaps I am missing something about what your constraints are, or what you're trying to achieve, but is there any reason you could not use a series of joined tables indicating parent-child relationships?  The following example follows that in your previous posts.  Note that this approach (as given) will not work if branches stemming from the same node are different lengths.  That is, if you have costs associated with "Transportation.Bicycle.Gear", you could not also have a category "Transportation.Bicycle.Gear.Chain_ring".  (To add the latter category, you'd have to put costs from the former under something like "Transportation.Bicycle.Gear.General" -- or modify the approach.)  However, lengthening the "Alcohol" branches, e.g., by tacking on a level5 table would be easy.  Notice that level3 and level4 are not true look-up tables, since they may contain duplicate cat values.

If I'm off base, by all means specify just how.

Regards,

Bryan

--------------------------------------------------

CREATE TABLE level1 (

  cat   text  PRIMARY KEY

);

CREATE TABLE level2 (

   cat   text   PRIMARY KEY,

   parent   text   REFERENCES level1(cat)

);

CREATE TABLE level3 (

   cat   text,

   parent   text   REFERENCES level2(cat),

   cost   numeric(6,2)

);

CREATE TABLE level4 (

   cat   text,

   parent   text,

   cost   numeric(6,2)

);

INSERT INTO level1

  VALUES ('Transportation'),

         ('Groceries');

INSERT INTO level2

  VALUES ('Auto', 'Transportation'),

         ('Bicycle', 'Transportation'),

         ('Fares', 'Transportation'),

         ('Food', 'Groceries'),

         ('Beverages', 'Groceries');

INSERT INTO level3

  VALUES ('Fuel', 'Auto', 50.00),

         ('Maintenance', 'Auto', 30.00),

         ('Fuel', 'Auto', 25.00), 

         ('Gear', 'Bicycle', 40.00),

         ('Gear', 'Bicycle', 20.00),

         ('Bus', 'Fares', 10.00),

         ('Train', 'Fares', 5.00),

         ('Beverages', 'Food', 30.00),

         ('Fruit_Veg', 'Food', 40.00),

         ('Meat_Fish', 'Food', 80.00),

         ('Grains_Cereals', 'Food', 30.00), 

         ('Alcohol', 'Beverages', NULL),

         ('Juice', 'Beverages', 45.00), 

         ('Other', 'Beverages', 15.00); 

INSERT INTO level4

  VALUES ('Beer', 'Alcohol', 25.00),

         ('Spirits', 'Alcohol', 10.00),

         ('Wine', 'Alcohol', 50.00);

CREATE VIEW all_cats AS (

SELECT a.cat AS level4,

       b.cat AS level3,

       c.cat AS level2,

       d.cat AS level1,

       CASE WHEN a.cost IS NULL THEN 0

            WHEN a.cost IS NOT NULL THEN a.cost

                 END

       + CASE WHEN b.cost IS NULL THEN 0

              WHEN b.cost IS NOT NULL THEN b.cost

         END AS cost

  FROM level4 a

    FULL JOIN

    level3 b

    ON (a.parent = b.cat)

      FULL JOIN

      level2 c

      ON (b.parent = c.cat)

        FULL JOIN

        level1 d

        ON (c.parent = d.cat)

  ORDER BY level1, level2, level3, level4

);

SELECT * FROM all_cats;

     level4  |     level3     |  level2   |     level1     | cost  

    ---------+----------------+-----------+----------------+-------

     Beer    | Alcohol        | Beverages | Groceries      | 25.00

     Spirits | Alcohol        | Beverages | Groceries      | 10.00

     Wine    | Alcohol        | Beverages | Groceries      | 50.00

             | Juice          | Beverages | Groceries      | 45.00

             | Other          | Beverages | Groceries      | 15.00

             | Beverages      | Food      | Groceries      | 30.00

             | Fruit_Veg      | Food      | Groceries      | 40.00

             | Grains_Cereals | Food      | Groceries      | 30.00

             | Meat_Fish      | Food      | Groceries      | 80.00

             | Fuel           | Auto      | Transportation | 50.00

             | Fuel           | Auto      | Transportation | 25.00

             | Maintenance    | Auto      | Transportation | 30.00

             | Gear           | Bicycle   | Transportation | 20.00

             | Gear           | Bicycle   | Transportation | 40.00

             | Bus            | Fares     | Transportation | 10.00

             | Train          | Fares     | Transportation |  5.00

    (16 rows)

SELECT level1,

       count(cost) AS num_branches,

       sum(cost) AS total_cost 

  FROM all_cats

  GROUP BY level1

  ORDER BY level1;

         level1     | num_branches | total_cost 

    ----------------+--------------+------------

     Groceries      |            9 |     325.00

     Transportation |            7 |     180.00

    (2 rows)

Quoth parrisdc@gmail.com (Don Parris):

>
> Is it possible to use spaces in the ltree path, like so:
> TOP.Groceries.Food.Herbs & Spices
>
> Or do the elements of the path have to use underscores and dashes?

From the docs:

|  A label is a sequence of alphanumeric characters and underscores (for
|  example, in C locale the characters A-Za-z0-9_ are allowed). Labels
|  must be less than 256 bytes long. [...]
|
| A label path is a sequence of zero or more labels separated by dots,
| for example L1.L2.L3, representing a path from the root of a
| hierarchical tree to a particular node. The length of a label path
| must be less than 65Kb, but keeping it under 2Kb is preferable.

If you need to store non-alphanumeric labels, one answer (as long as
they aren't too long) would be to use URL-encoding, like

    TOP.Groceries.Food.Herbs_20_26_20Spices

Of course, you would need to encode _ as well, and you would need to be
sure the labels weren't going to come out too long. Another alternative
would be to MD5 each label and use (say) the first 10 bytes of that MD5
in hex as the ltree label. (Annoyingly there's only one
non-alphanumeric, so you can't use base64.) If you were going to do that
you would need to consider the possibility of an attacker arranging a
hash collision: I don't know where you're labels come from, so I don't
know if this would be an issue.

Ben