Best way in MySQL or Rails to get AVG per day within a specific date range

Question

I\'m trying to make a graph in Rails, for example the avg sales amount per day for each day in a given date range

Say I have a products_sold model which has a \"sales_pr

Answer 1

Is there a reason (other than the date one already mentioned) why you wouldn't use the built-in group function capabilities in ActiveRecord? You seem to be concerned about "post-processing", which I don't think is really something to worry about.

You're in Rails, so you should probably be looking for a Rails solution first[1]. My first thought would be to do something like

Product.average(:sales_price, :group => "DATE(created_at)", :conditions => ["merchant_id=?", 1])

which ActiveRecord turned into pretty much the SQL you described. Assuming there's a declared has_many association between Merchant and Product, then you'd probably be better using that, so something like:

ave_prices = Merchant.find(1).products.average(:sales_price, :group => "DATE(created_at)")

(I'm hoping that your description of the model as "products_sold" is some kind of transcription error, btw - if not, you're somewhat off-message with your class naming!)

After all that, you're back where you started, but you got there in a more conventional Rails way (and Rails really values conventions!). Now we need to fill in the gaps.

I'll assume you know your date range, let's say it's defined as all dates from from_date to to_date.

date_aves = (from_date..to_date).map{|dt| [dt, 0]}

That builds the complete list of dates as an array. We don't need the dates where we got an average:

ave_price_dates = ave_prices.collect{|ave_price| ave_price[0]} # build an array of dates
date_aves.delete_if { |dt| ave_price.dates.index(dt[0]) } # remove zero entries for dates retrieved from DB
date_aves.concat(ave_prices)     # add the query results
date_aves.sort_by{|ave| ave[0] } # sort by date

That lot looks a bit cluttered to me: I think it could be terser and cleaner. I'd investigate building a Hash or Struct rather than staying in arrays.

---

[1] I'm not saying don't use SQL - situations do occur where ActiveRecord can't generate the most efficient query and you fall back on find_by_sql. That's fine, it's supposed to be like that, but I think you should try to use it only as a last resort.

Answer 2

To dry up a bit:

ave_prices = Merchant.find(1).products.average(:sales_price, :group => "DATE(created_at)")
date_aves = (from_date..to_date).map{|dt| [dt, ave_prices[dt.strftime "%Y-%m-%d"] || 0]}

Answer 3

Does MySQL have set-returning functions? I.e. functions that return different values on each row of a query? As an example from PostgreSQL, you can do:

select 'foo', generate_series(3, 5);

This will produce a result set consisting of 2 columns and 3 rows, where the left column contains 'foo' on each row and the right column contains 3, 4 and 5.

So, assuming you have an equivalent of generate_series() in MySQL, and subqueries: What you need is a LEFT OUTER JOIN from this function to the query that you already have. That will ensure you see each date appear in the output:

SELECT
    avg(sales_price) as avg,
    DATE_FORMAT(the_date, '%m-%d-%Y') as date
FROM (select cast('2008-JAN-01' as date) + generate_series(0, 364) as the_date) date_range
LEFT OUTER JOIN products_sold on (the_date = created_at)
WHERE merchant_id = 1
GROUP BY date;

You may need to fiddle with this a bit to get the syntax right for MySQL.

Answer 4

For any such query, you will need to find a mechanism to generate a table with one row for each date that you want to report on. Then you will do an outer join of that table with the data table you are analyzing. You may also have to play with NVL or COALESCE to convert nulls into zeroes.

The hard part is working out how to generate the (temporary) table that contains the list of dates for the range you need to analyze. That is DBMS-specific.

Your idea of mapping date/time values to a single date is spot on, though. You'd need to pull a similar trick - mapping all the dates to an ISO 8601 date format like 2009-W01 for week 01 - if you wanted to analyze weekly sales.

Also, you would do better to map your DATE format to 2009-01-08 notation because then you can sort in date order using a plain character sort.