I have two tables, one is a table of employee names, 176 records. The other is a table (with duplicates) of employee names (same format) and their locations (7943 rows).
From this answer i deduced i needed a left join to give me the rows from Table A only.
I was hoping the below would give me the original 176 rows back from Table A, each column with a value for location from Table B, else blank if not available, however it gives me 7601 rows which i cannot for the life of me understand:
SELECT e.[UniqueName], l.[location]
FROM [Employees] as e
left join Locations as l
on e.[UniqueName] = l.[UniqueName]
Even using a group by (which I'm not sure why this would be necessary given that I am asking only for whats in Table A) gives 172 rows even though each name in the Employees table is unique!
The table Locations contains more than 1 locations for each employee and this is why you get so many rows in the results.
If you want just 1 location and it does not matter which 1 then add aggregation to your query:
SELECT e.[UniqueName], MAX(l.[location]) AS location
FROM [Employees] as e
LEFT JOIN Locations as l
ON e.[UniqueName] = l.[UniqueName]
GROUP BY e.[UniqueName]
You can use MIN() instead of MAX().
I was hoping the below would give me the original 176 rows back from
Table A, each column with a value for location from Table B, else
blank if not available, however it gives me 7601 rows which I cannot
for the life of me understand...
Whilst a left join will always return all records from the dataset on the lefthand side of the join, the number of records returned by the query will depend upon the number of possible pairings between the two datasets, which (for a left join) will always be greater than or equal to the number of records in the dataset to the left of the join.
For your example, consider the following two datasets:
Employees
+------------+
| UniqueName |
+------------+
| Alice |
| Bob |
| Charlie |
+------------+
Locations
+------------+----------+
| UniqueName | Location |
+------------+----------+
| Alice | London |
| Bob | Berlin |
| Bob | New York |
| Bob | Paris |
+------------+----------+
Evaluating the query:
select
e.[uniquename], l.[location]
from
[employees] as e left join locations as l
on e.[uniquename] = l.[uniquename]
Will cause the records to be paired up in the following manner:
And will therefore return the result:
+------------+----------+
| uniquename | location |
+------------+----------+
| Alice | London |
| Bob | Berlin |
| Bob | New York |
| Bob | Paris |
| Charlie | |
+------------+----------+
You can use a correlated subquery:
SELECT e.[UniqueName],
(SELECT TOP 1 l.[location]
FROM locations as l
WHERE e.[UniqueName] = l.[UniqueName]
) as location
FROM [Employees] as e;
Note: There is no ORDER BY so this returns an arbitrary location.
If location can be duplicated for a given UniqueName, you will get an error. To solve that, you can use an aggregation functions:
SELECT e.[UniqueName],
(SELECT MAX(l.[location])
FROM locations as l
WHERE e.[UniqueName] = l.[UniqueName]
) as location
FROM [Employees] as e;
Related
I just started studying SQL and this is a demo given by the teacher in an online course and it works fine. The statement is looking for "students such that number of other students with same GPA is equal to number of other students with same sizeHS":
select *
from Student S1
where (
select count(*)
from Student S2
where S2.sID <> S1.sID and S2.GPA = S1.GPA
) = (
select count(*)
from Student S2
where S2.sID <> S1.sID and S2.sizeHS = S1.sizeHS
);
It seems that in this where clause, we're comparing two relations (because the result of a subquery is a relation), but most of the time we are comparing attributes(as far as I've seen).
So I'm thinking about whether there are requirements for how many attributes, and how many tuples, the RELATION should contain when comparing two RELATIONS. If not, how do we compare two RELATIONS when there're multiple attributes or multiple tuples and what do we get for result?
Note:
Student relation has 4 attributes: sID, sName, GPA, sizeHS. And here's the data:
+-----+--------+-----+--------+
| sID | sName | GPA | sizeHS |
+-----+--------+-----+--------+
| 123 | Amy | 3.9 | 1000 |
| 234 | Bob | 3.6 | 1500 |
| 345 | Craig | 3.5 | 500 |
| 456 | Doris | 3.9 | 1000 |
| 567 | Edward | 2.9 | 2000 |
| 678 | Fay | 3.8 | 200 |
| 789 | Gary | 3.4 | 800 |
| 987 | Helen | 3.7 | 800 |
| 876 | Irene | 3.9 | 400 |
| 765 | Jay | 2.9 | 1500 |
| 654 | Amy | 3.9 | 1000 |
| 543 | Craig | 3.4 | 2000 |
+-----+--------+-----+--------+
and the result of this query is:
+-----+--------+-----+---------+
| sID | sName | GPA | sizeHS |
+-----+--------+-----+---------+
| 345 | Craig | 3.5 | 500 |
| 567 | Edward | 2.9 | 2000 |
| 678 | Fay | 3.8 | 200 |
| 789 | Gary | 3.4 | 800 |
| 765 | Jay | 2.9 | 1500 |
| 543 | Craig | 3.4 | 2000 |
+-----+--------+-----+---------+
because the result of a subquery is a relation
Relation is the scientific name for what we call a table in a database and I like the name "table" much better than "relation". A table is easy to imagine. We know them from our school time schedule for instance. Yes, we relate things here inside a table (day and time and the subject taught in school), but we can also relate tables to tables (pupils' timetables with the table of class rooms, the overall subject schedule, and the teacher's timetables). As such, tables in an RDBMS are also related to each other (hence the name relational database management system). I find the name relation for a table quite confusing (and many people use the word "relation" to describe the relations between tables instead).
So, yes, a query result itself is again a table ("relation"). And from tables we can of course select:
select * from (select * from b) as subq;
And then there are scalar queries that return exactly one row and one column. select count(*) from b is such a query. While this is still a table we can select from
select * from (select count(*) as cnt from b) as subq;
we can even use them where we usually have single values, e.g. in the select clause:
select a.*, (select count(*) from b) as cnt from a;
In your query you have two scalar subqueries in your where clause.
With subqueries there is another distinction to make: we have correlated and non-correlated subqueries. The last query I have just shown contains a non-correlated subquery. It selects the count of b rows for every single result row, no matter what that row contains elsewise. A correlated subquery on the other hand may look like this:
select a.*, (select count(*) from b where b.x = a.y) as cnt from a;
Here, the subquery is related to the main table. For every result row we look up the count of b rows matching the a row we are displaying via where b.x = a.y, so the count is different from row to row (but we'd get the same count for a rows sharing the same y value).
Your subqueries are also correlated. As with the select clause, the where clause deals with one row at a time (in order to keep or dismiss it). So we look at one student S1 at a time. For this student we count other students (S2, where S2.sID <> S1.sID) who have the same GPA (and S2.GPA = S1.GPA) and count other students who have the same sizeHS. We only keep students (S1) where there are exactly as many other students with the same GPA as there are with the same sizeHS.
UPDATE
As do dealing with multiple tuples as in
select *
from Student S1
where (
select count(*), avg(grade)
from Student S2
where S2.sID <> S1.sID and S2.GPA = S1.GPA
) = (
select count(*), avg(grade)
from Student S2
where S2.sID <> S1.sID and S2.sizeHS = S1.sizeHS
);
this is possible in some DBMS, but not in SQL Server. SQL Server doesn't know tuples.
But there are other means to achieve the same. You could just add two subqueries:
select * from student s1
where (...) = (...) -- compare counts here
and (...) = (...) -- compare averages here
Or get the data in the FROM clause and then deal with it. E.g.:
select *
from Student S1
cross apply
(
select count(*) as cnt, avg(grade) as avg_grade
from Student S2
where S2.sID <> S1.sID and S2.GPA = S1.GPA
) sx
cross apply
(
select count(*) as cnt, avg(grade) as avg_grade
from Student S2
where S2.sID <> S1.sID and S2.sizeHS = S1.sizeHS
) sy
where sx.cnt = sy.cnt and sx.avg_grade = sy.avg_grade;
There are relational operations:
The intersection operator produces the set of tuples that two
relations share in common. Intersection is implemented in SQL in the
form of the INTERSECT operator.
The difference operator acts on two relations and produces the set of tuples from the first relation that do not exist in the second relation. Difference is implemented in SQL in the form of the EXCEPT or MINUS operator.
So, in the context of SQL Server, for example, you can do:
SELECT *
FROM R1
EXCEPT
SELECT *
FROM R2
to get rows in R1 not included in R2 and the reverse - to get all differences.
Of course, the attributes must be the same - if not, you need to explicit set the attributes in the SELECT.
I Have a query that finds a table, here's an example one.
Name |Age |Hair |Happy | Sad |
Jon | 15 | Black |NULL | NULL|
Kyle | 18 |Blonde |YES |NULL |
Brad | 17 | Blue |NULL |YES |
Name and age come from one table in a database, hair color comes from a second which is joined, and happy and sad come from a third table.My goal would be to make the first line of the chart like this:
Name |Age |Hair |Happy |Sad |
Jon | 15 |Black |Yes |Yes |
Basically I want to get rid of the rows under the first and get the non NULL data joined to the right. The problem is that there is no column where the Yes values are on the Jon row, so I have no idea how to get them there. Any suggestions?
PS. With the data I am using I can't just put a 'YES' in the 'Jon' row and call it a day, I would need to find the specific value from the lower rows and somehow get that value in the boxes that are NULL.
Do you just want COALESCE()?
COALESCE(Happy, 'Yes') as happy
COALESCE() replaces a NULL value with another value.
If you want to join on a NULL value work with nested selects. The inner select gets an Id for NULLs, the outer select joins
select COALESCE(x.Happy, yn_table.description) as happy, ...
from
(select
t1.Happy,
CASE WHEN t1.Happy is null THEN 1 END as happy_id
from t1 ...) x
left join yn_table
on x.xhappy_id = yn_table.id
If you apply an ORDER BY to the query, you can then select the first row relative to this order with WHERE rownum = 1. If you don't apply an ORDER BY, then the order is random.
After reading your new comment...
the sense is that in my real data the yes under the other names will be a number of a piece of equipment. I want the numbers of the equipment in one row instead of having like 8 rows with only 4 ' yes' values and the rest null.
... I come to the conclusion that this a XY problem.
You are asking about a detail you think will solve your problem, instead of explaining the problem and asking how to solve it.
If you want to store several pieces of equipment per person, you need three tables.
You need a Person table, an Article table and a junction table relating articles to persons to equip them. Let's call this table Equipment.
Person
------
PersonId (Primary Key)
Name
optional attributes like age, hair color
Article
-------
ArticleId (Primary Key)
Description
optional attributes like weight, color etc.
Equipment
---------
PersonId (Primary Key, Foreign Key to table Person)
ArticleId (Primary Key, Foreign Key to table Article)
Quantity (optional, if each person can have only one of each article, we don't need this)
Let's say we have
Person: PersonId | Name
1 | Jon
2 | Kyle
3 | Brad
Article: ArticleId | Description
1 | Hat
2 | Bottle
3 | Bag
4 | Camera
5 | Shoes
Equipment: PersonId | ArticleId | Quantity
1 | 1 | 1
1 | 4 | 1
1 | 5 | 1
2 | 3 | 2
2 | 4 | 1
Now Jon has a hat, a camera and shoes. Kyle has 2 bags and one camera. Brad has nothing.
You can query the persons and their equipment like this
SELECT
p.PersonId, p.Name, a.ArticleId, a.Description AS Equipment, e.Quantity
FROM
Person p
LEFT JOIN Equipment e
ON p.PersonId = e.PersonId
LEFT JOIN Article a
ON e.ArticleId = a.ArticleId
ORDER BY p.Name, a.Description
The result will be
PersonId | Name | ArticleId | Equipment | Quantity
---------+------+-----------+-----------+---------
3 | Brad | NULL | NULL | NULL
1 | Jon | 4 | Camera | 1
1 | Jon | 1 | Hat | 1
1 | Jon | 5 | Shoes | 1
2 | Kyle | 3 | Bag | 2
2 | Kyle | 4 | Camera | 1
See example: http://sqlfiddle.com/#!4/7e05d/2/0
Since you tagged the question with the oracle tag, you could just use NVL(), which allows you to specify a value that would replace a NULL value in the column you select from.
Assuming that you want the 1st row because it contains the smallest age:
- wrap your query inside a CTE
- in another CTE get the 1st row of the query
- in another CTE get the max values of Happy and Sad of your query (for your sample data they both are 'YES')
- cross join the last 2 CTEs.
with
cte as (
<your query here>
),
firstrow as (
select name, age, hair from cte
order by age
fetch first row only
),
maxs as (
select max(happy) happy, max(sad) sad
from cte
)
select f.*, m.*
from firstrow f cross join maxs m
You can try this:
SELECT A.Name,
A.Age,
B.Hair,
C.Happy,
C.Sad
FROM A
INNER JOIN B
ON A.Name = B.Name
INNER JOIN C
ON A.Name = B.Name
(Assuming that Name is the key columns in the 3 tables)
I'm trying to better understand relational algebra and am having trouble solving the following type of question:
Suppose there is a column A (Department), a column B (Employees) and a column C (Managers). How can I find all of the departments who only have one manager for all of their employees? An example is provided below:
Department | Employees | Managers
-------------+-------------+----------
A | John | Bob
A | Sue | Sam
B | Jim | Don
B | Alex | Don
C | Jason | Xie
C | Greg | Xie
In this table, the result I should get are all tuples containing departments B and C because all of their employees are managed by the same person (Don and Xie respectively). Department A however, would not be returned because it's employees have multiple managers.
Any help or pointers would be appreciated.
Such problems usually call for a self-join.
Joining the relation onto itself on Department, then filtering out the tuples where the Managers are equal would yield us all the unwanted tuples, which we can just subtract from the original relations.
Here's how I'd do it:
First we make a copy of table T, and call it T2, then take a cross product of T and T2. From the result we select all the rows where T1.Manager /= T2.Manager but T1.Department=T2.Department, yielding us these tuples:
T1.Department | T1.Employees| T1.Managers | T2.Managers | T2.Employees | T2.Department
--------------+-------------+-------------+-------------+--------------+--------------
A | John | Bob | Sam | Sue | A
A | Sue | Sam | Bob | John | A
Departments A and B aren't present because their T1.Manager always equals T2.Manager.
Then we just subtract this result the original set to get the answer.
If your RDBMS supports common table expressions:
with C as (
select department, manager, count(*) as cnt
from A
group by department, manager
),
B as (
select department, count(*) as cnt
from A group by department
)
select A.*
from A
join C on A.department = C.department
join B on A.department = B.department
where B.cnt = C.cnt;
I'm trying to use a join on three tables query I found in another post (post #5 here). When I try to use this in the SQL tab of one of my tables in phpMyAdmin, it gives me an error:
#1066 - Not unique table/alias: 'm'
The exact query I'm trying to use is:
select r.*,m.SkuAbbr, v.VoucherNbr from arrc_RedeemActivity r, arrc_Merchant m, arrc_Voucher v
LEFT OUTER JOIN arrc_Merchant m ON (r.MerchantID = m.MerchantID)
LEFT OUTER JOIN arrc_Voucher v ON (r.VoucherID = v.VoucherID)
I'm not entirely certain it will do what I need it to do or that I'm using the right kind of join (my grasp of SQL is pretty limited at this point), but I was hoping to at least see what it produced.
(What I'm trying to do, if anyone cares to assist, is get all columns from arrc_RedeemActivity, plus SkuAbbr from arrc_Merchant where the merchant IDs match in those two tables, plus VoucherNbr from arrc_Voucher where VoucherIDs match in those two tables.)
Edited to add table samples
Table arrc_RedeemActivity
RedeemID | VoucherID | MerchantID | RedeemAmt
----------------------------------------------
1 | 2 | 3 | 25
2 | 6 | 5 | 50
Table arrc_Merchant
MerchantID | SkuAbbr
---------------------
3 | abc
5 | def
Table arrc_Voucher
VoucherID | VoucherNbr
-----------------------
2 | 12345
6 | 23456
So ideally, what I'd like to get back would be:
RedeemID | VoucherID | MerchantID | RedeemAmt | SkuAbbr | VoucherNbr
-----------------------------------------------------------------------
1 | 2 | 3 | 25 | abc | 12345
2 | 2 | 5 | 50 | def | 23456
The problem was you had duplicate table references - which would work, except for that this included table aliasing.
If you want to only see rows where there are supporting records in both tables, use:
SELECT r.*,
m.SkuAbbr,
v.VoucherNbr
FROM arrc_RedeemActivity r
JOIN arrc_Merchant m ON m.merchantid = r.merchantid
JOIN arrc_Voucher v ON v.voucherid = r.voucherid
This will show NULL for the m and v references that don't have a match based on the JOIN criteria:
SELECT r.*,
m.SkuAbbr,
v.VoucherNbr
FROM arrc_RedeemActivity r
LEFT JOIN arrc_Merchant m ON m.merchantid = r.merchantid
LEFT JOIN arrc_Voucher v ON v.voucherid = r.voucherid
I am trying to figure out how to use multiple left outer joins to calculate average scores and number of cards. I have the following schema and test data. Each deck has 0 or more scores and 0 or more cards. I need to calculate an average score and card count for each deck. I'm using mysql for convenience, I eventually want this to run on sqlite on an Android phone.
mysql> select * from deck;
+----+-------+
| id | name |
+----+-------+
| 1 | one |
| 2 | two |
| 3 | three |
+----+-------+
mysql> select * from score;
+---------+-------+---------------------+--------+
| scoreId | value | date | deckId |
+---------+-------+---------------------+--------+
| 1 | 6.58 | 2009-10-05 20:54:52 | 1 |
| 2 | 7 | 2009-10-05 20:54:58 | 1 |
| 3 | 4.67 | 2009-10-05 20:55:04 | 1 |
| 4 | 7 | 2009-10-05 20:57:38 | 2 |
| 5 | 7 | 2009-10-05 20:57:41 | 2 |
+---------+-------+---------------------+--------+
mysql> select * from card;
+--------+-------+------+--------+
| cardId | front | back | deckId |
+--------+-------+------+--------+
| 1 | fron | back | 2 |
| 2 | fron | back | 1 |
| 3 | f1 | b2 | 1 |
+--------+-------+------+--------+
I run the following query...
mysql> select deck.name, sum(score.value)/count(score.value) "Ave",
-> count(card.front) "Count"
-> from deck
-> left outer join score on deck.id=score.deckId
-> left outer join card on deck.id=card.deckId
-> group by deck.id;
+-------+-----------------+-------+
| name | Ave | Count |
+-------+-----------------+-------+
| one | 6.0833333333333 | 6 |
| two | 7 | 2 |
| three | NULL | 0 |
+-------+-----------------+-------+
... and I get the right answer for the average, but the wrong answer for the number of cards. Can someone tell me what I am doing wrong before I pull my hair out?
Thanks!
John
It's running what you're asking--it's joining card 2 and 3 to scores 1, 2, and 3--creating a count of 6 (2 * 3). In card 1's case, it joins to scores 4 and 5, creating a count of 2 (1 * 2).
If you just want a count of cards, like you're currently doing, COUNT(Distinct Card.CardId).
select deck.name, coalesce(x.ave,0) as ave, count(card.*) as count -- card.* makes the intent more clear, i.e. to counting card itself, not the field. but do not do count(*), will make the result wrong
from deck
left join -- flatten the average result rows first
(
select deckId,sum(value)/count(*) as ave -- count the number of rows, not count the column name value. intent is more clear
from score
group by deckId
) as x on x.deckId = deck.id
left outer join card on card.deckId = deck.id -- then join the flattened results to cards
group by deck.id, x.ave, deck.name
order by deck.id
[EDIT]
sql has built-in average function, just use this:
select deckId, avg(value) as ave
from score
group by deckId
What's going wrong is that you're creating a Cartesian product between score and card.
Here's how it works: when you join deck to score, you may have multiple rows match. Then each of these multiple rows is joined to all of the matching rows in card. There's no condition preventing that from happening, and the default join behavior when no condition restricts it is to join all rows in one table to all rows in another table.
To see it in action, try this query, without the group by:
select *
from deck
left outer join score on deck.id=score.deckId
left outer join card on deck.id=card.deckId;
You'll see a lot of repeated data in the columns that come from score and card. When you calculate the AVG() over data that has repeats in it, the redundant values magically disappear (as long as the values are repeated uniformly). But when you COUNT() or SUM() them, the totals are way off.
There may be remedies for inadvertent Cartesian products. In your case, you can use COUNT(DISTINCT) to compensate:
select deck.name, avg(score.value) "Ave", count(DISTINCT card.front) "Count"
from deck
left outer join score on deck.id=score.deckId
left outer join card on deck.id=card.deckId
group by deck.id;
This solution doesn't solve all cases of inadvertent Cartesian products. The more general-purpose solution is to break it up into two separate queries:
select deck.name, avg(score.value) "Ave"
from deck
left outer join score on deck.id=score.deckId
group by deck.id;
select deck.name, count(card.front) "Count"
from deck
left outer join card on deck.id=card.deckId
group by deck.id;
Not every task in database programming must be done in a single query. It can even be more efficient (as well as simpler, easier to modify, and less error-prone) to use individual queries when you need multiple statistics.
Using left joins isn't a good approach, in my opinion. Here's a standard SQL query for the result you want.
select
name,
(select avg(value) from score where score.deckId = deck.id) as Ave,
(select count(*) from card where card.deckId = deck.id) as "Count"
from deck;