count first time occurrence of foreign keys - php

Question

Besides the tags, I would like to solve this on query, if possible.
I have this table
activity_type | value | date | company_id
network.new | 1 | 2011-10-08 | 1
members.count | 3 | 2011-10-08 | 1
network.new | 1 | 2011-10-10 | 2
network.new | 1 | 2011-10-11 | 3
members.count | 4 | 2011-10-11 | 2
That's basically a log activity.
'network.new' activity occur only
once per company_id
'members_count' activity occur only after
'network.new' appears per company_id and can appear once per day per
company_id.
I need to make a line graph that the X axis is the date, the Y axis is the quantity of two things:
How many company_ids have members each day of activity for the first
time (That is the one that is giving me a hard time);
How many have the network.new activity and only that activity
for each given day.
All queries I tried gave me false-positive lists, mostly because it counts company_ids that have the 'members_count' activity every day.
I wish, if possible, to create a query that give me date, first_time_members, new_company columns for create view purposes.
I hope my question was clear enough, and not silly because I couldn't find it anything that looks close to my problem anywhere.
[EDIT]
Since my english is really poor, I couldn't make myself clear I'm going to try explain a litle more:
My client have a network of companies and he wishes to learn how many
companies join the network day by day but there's a catch: A company
when sign in for the network, it is only considered a completed
one when it is also had registered members. So he wants to know, how many
companies make a 'incomplete' sign up and how many make a 'complete'
sign up.
Mr Ollie Jones put me in the right direction, I think I can use what he tough me, but it is not right there yet.
Thank ollie Jones for your answer by the way. Answers like yours made me love this site.

Answer 1

I'm going to stick my neck out and guess what you want. You're asking for "How many company_ids have members each day of activity for the first time". With respect, this is a very hard statement to understand.
I think you mean this: for each day, how many company_id values appear for the very first time in a network.new activity type, and how many of those are accompanied by nonzero members.count item in that same day, and how many are not?
Here's what you do:
First think of a query that will give the very first date for each company appearing in your system. Try this.
SELECT MIN(date) networknewdate, company_id
FROM table
WHERE activity_type = 'network.new'
GROUP BY company_id
This yields a virtual table of networknewdate, company_id.
Next, you need a query that will give the first date a members.count item turns up for each company.
SELECT MIN(date) memberscountdate, company_id
FROM table
WHERE activity_type = 'members.count'
GROUP BY date
OK, now we have two nice virtual tables each with at most one row for each company_id value. Let's join them, driving the join off the first (network.new) value.
SELECT a.networknewdate,
a.company_id,
IFNULL(b.members_present, 'no') members
FROM (
SELECT MIN(date) networknewdate, company_id
FROM table
WHERE activity_type = 'network.new'
GROUP BY company_id
) a
LEFT JOIN (
SELECT MIN(date) memberscountdate, company_id, 'yes' members_present
FROM table
WHERE activity_type = 'members.count'
GROUP BY date
) b ON (a.networknewdate = b.memberscountdate and a.company_id = b.company_id)
This will return three columns: a date, a company_id, and 'yes' or 'no' saying whether there was a first members.count record on the same day as the first network.new record for each company_id.
Now you need to summarize this whole thing so you get one record per day, with the number of 'yes' and the number of 'no' items listed. Here we go.
The number of 'yes' records by day.
SELECT networknewdate, count(*) yesrecords
FROM (
SELECT a.networknewdate,
a.company_id,
IFNULL(b.members_present, 'no') members
FROM (
SELECT MIN(date) networknewdate, company_id
FROM table
WHERE activity_type = 'network.new'
GROUP BY company_id
) a
LEFT JOIN (
SELECT MIN(date) memberscountdate, company_id, 'yes' members_present
FROM table
WHERE activity_type = 'members.count'
GROUP BY date
) b ON (a.networknewdate = b.memberscountdate and a.company_id = b.company_id)
) r
WHERE r.members = 'yes'
GROUP BY networknewdate
The number of no records by date is a similar query. Then you need to left join those two queries together on the networknewdate so you get a table of date, yesrecords, norecords. I'm going to leave this as a cut 'n paste exercise for you. It's more than twice as long as the query I wrote ending in GROUP BY networknewdate.
Welcome to SQL that implements real world business logic! I think the take-home lesson on this question is that you're asking for a result that's actually quite hard to specify. Once you specify exactly what you want, writing a query to get it is tedious and repetitive but not hard.
Another little hint. It may make sense for you to create some views so your queries aren't so enormous.

Answer 2

So using the same approach Ollie jones showed me I figure it out:
First I need a list of dates where 'members_count' OR 'network.new' happens
SELECT date as current_date
FROM activity_log ld
WHERE `activity_type` in ('members_count', 'network.new')
GROUP BY date
ORDER BY date
Them I left Join with a list of first date companies appears
SELECT MIN(date) AS new_date, company_id
FROM activity_log
WHERE activity_type = 'network.new'
GROUP BY company_id
ORDER BY date
Also Left Join first time a company count members
SELECT min(date) as members_count_date, company_id
FROM `activity_networks` WHERE `activity_type` = 'network.daily.members_count'
GROUP BY company_id
ORDER BY date
Then a Make a distinct count of new companies and companies that count members for the first time, group by date. Then I have this:
SELECT DATE(FROM_UNIXTIME(ld.date)) as curr_date,
COUNT(DISTINCT(new_co)) as new_co,
COUNT(DISTINCT(complete_co)) as complete
FROM activity_log ld
LEFT JOIN (SELECT date AS new_date, company_id as new_co
FROM activity_networks
WHERE activity_type = 'network.new'
GROUP BY company_id
ORDER BY date) nd ON (ld.date=nd.new_date)
LEFT JOIN (SELECT min(date) as members_count_date, company_id as complete_co
FROM `activity_log` WHERE `activity_type` = 'members_count'
GROUP BY company_id
ORDER BY date) mcd ON (mcd.members_count_date=ld.date)
WHERE `activity_type` in ('members_count', 'network.new')
GROUP BY DATE(FROM_UNIXTIME(ld.date))
ORDER BY ld.date
The distinct function was crucial because the counting wasn't doing right without it. It is not perfect. The column I named 'new_co' should bring only incomplete registrations (incomplete means, new register with out members linked to a company), but still the information can be useful.