MySQL query with multiple random values but sum always within a range - php

Question

I have a table of store items with their price.
I'm trying to write a mysql query that pulls a number of items (between 3 and 6) at RANDOM, with the TOTAL value of all items within $20 of a value chosen by the user.
Any idea's on how to accomplish this?
Thanks in advance!
EDIT*** so far here is what I have. The big issue is that sum(price) takes the sum of ALL items. The secondary issue is having the "LIMIT" be random, but I can eventually have php pick a random number between 3 and 6 prior to running the query.
SELECT item,price,sum(price)
FROM items
WHERE sum(price) BETWEEN ($value-10) AND ($value+10)
ORDER BY rand() LIMIT 6

Answer 1

I can't think of a good way to do this in an SQL query without joining your items table on itself multiple times, which would lead to combinatorial explosion as the number of items in the table grows.
I've worked up a solution in PHP that breaks your items into price groups. Consider the following table:
+----+--------------------+-------+
| id | item | price |
+----+--------------------+-------+
| 1 | apple | 10.5 |
| 2 | banana | 1.85 |
| 3 | carrot | 16.22 |
| 4 | donut | 13.33 |
| 5 | eclair | 18.85 |
| 6 | froyo | 26.99 |
| 7 | gingerbread | 12.15 |
| 8 | honecomb | 50.68 |
| 9 | ice-cream-sandwich | 2.44 |
| 10 | jellybean | 2.45 |
| 11 | kitkat | 2.46 |
| 12 | lollipop | 42.42 |
+----+--------------------+-------+
http://sqlfiddle.com/#!9/0d815
First, break the items into Price Groups based on the random number of items (between 3 and 6 in your case). The Price Group Increment will be determined by the variance in price ($20.00) divided by the number of items being selected. This ensures that you will not go outside of your variance. Here is an example for a group of 4 items:
PRICE_GROUP_INCREMENT = VARIANCE / NUMBER_ITEMS
PRICE_GROUP_INCREMENT = 20 / 4 = 5
SELECT Count(`id`) AS `item_count`,
Round(`price` / 5) `price_group`
FROM `items`
WHERE `price` <= 35
GROUP BY `price_group`
ORDER BY `price_group` ASC;
Result set:
+------------+-------------+
| item_count | price_group |
+------------+-------------+
| 4 | 0 |
| 2 | 2 |
| 2 | 3 |
| 1 | 4 |
+------------+-------------+
Next, we can search through the result set to find a combination of price groups that equal the target price group. The target price group is determined by your target price divided by the price group increment. Using our example above, let's try to find 4 items that add up to $35.00 with a $20.00 variance.
TARGET_PRICE_GROUP = round(TARGET_PRICE / PRICE_GROUP_INCREMENT)
TARGET_PRICE_GROUP = round($35.00 / $5.00) = 7
Searching through the result set, we have can get to a target price group of 7 with these groups of 4 items:
SELECT `items`.* FROM `items` WHERE ROUND(`price`/5) = 0 ORDER BY rand() ASC LIMIT 2;
SELECT `items`.* FROM `items` WHERE ROUND(`price`/5) = 4 ORDER BY rand() ASC LIMIT 1;
SELECT `items`.* FROM `items` WHERE ROUND(`price`/5) = 3 ORDER BY rand() ASC LIMIT 1;
or
SELECT `items`.* FROM `items` WHERE ROUND(`price`/5) = 0 ORDER BY rand() ASC LIMIT 1;
SELECT `items`.* FROM `items` WHERE ROUND(`price`/5) = 3 ORDER BY rand() ASC LIMIT 1;
SELECT `items`.* FROM `items` WHERE ROUND(`price`/5) = 2 ORDER BY rand() ASC LIMIT 2;
To speed up finding a random, suitable combination of queries, I wrote a recursive function that randomly weights each price group based on the number of items in it, then sorts it. This speeds things up because the function returns as soon as it finds the first solution. Here's the full PHP script:
<?php
function rand_weighted($weight, $total){
return (float)mt_rand()*(float)$weight/((float)mt_getrandmax()*(float)$total);
};
//you can change these
$targetPrice = 35.00;
$numProducts = rand(3,6);
$maxVariance = 20.00;
$priceGroupIncrement = $maxVariance / $numProducts;
$targetPriceGroupSum = (int)round($targetPrice/$priceGroupIncrement, 0);
$select = "SELECT COUNT(`id`) AS `item_count`, ROUND(`price`/{$priceGroupIncrement}) `price_group`";
$from = "FROM `items`";
$where = "WHERE `price` <= {$targetPrice}";
$groupBy = "GROUP BY `price_group`";
$orderBy = "ORDER BY `price_group` ASC"; //for readability of result set, not necessary
$sql = "{$select} {$from} {$where} {$groupBy} {$orderBy}";
echo "SQL for price groups:\n{$sql};\n\n";
//run your query here and get the result set
//here is a sample result set
//this assumes $targetPrice = 35.00, $numProducts=4, and $maxVariance=20.00
$numProducts = 4;
$priceGroupIncrement = 5;
$targetPriceGroupSum = 7;
$resultSet = array(
array('item_count'=>4, 'price_group'=>0),
array('item_count'=>2, 'price_group'=>2),
array('item_count'=>2, 'price_group'=>3),
array('item_count'=>1, 'price_group'=>4),
);
//end sample result set
$priceGroupItemCount = array();
$priceGroupWeight = array();
$total = 0;
//randomly weight price group based on how many items are in the group
foreach ($resultSet as $result){
$priceGroupItemCount[$result['price_group']] = $result['item_count'];
$total += $result['item_count'];
}
foreach ($resultSet as $result){
$priceGroupWeight[$result['price_group']] = rand_weighted($result['item_count'], $total);
}
//recursive anonymous function to find a match
$recurse = function($priceGroupWeight, $selection=array(), $priceGroupSum=0) use ($priceGroupItemCount, $total, $numProducts, $targetPriceGroupSum, &$recurse){
//sort by random weighted value
arsort($priceGroupWeight);
//iterate through each item in the $priceGroupWeight associative array
foreach ($priceGroupWeight as $priceGroup => $weight){
//copy variables so we can try a price group
$priceGroupWeightCopy = $priceGroupWeight;
$selectionCopy = $selection;
$priceGroupSumCopy = $priceGroupSum + $priceGroup;
//try to find a combination that adds up to the target price group
if (isset($selectionCopy[$priceGroup])){
$selectionCopy[$priceGroup]++;
} else {
$selectionCopy[$priceGroup] = 1;
}
$selectionCount = array_sum($selectionCopy);
if ($priceGroupSumCopy == $targetPriceGroupSum && $selectionCount == $numProducts) {
//we found a working solution!
return $selectionCopy;
} else if ($priceGroupSumCopy < $targetPriceGroupSum && $selectionCount < $numProducts) {
//remove the item from the price group
unset($priceGroupWeightCopy[$priceGroup]);
//if there is still remaining items in the group, add the adjusted weight back into the price group
$remainingInPriceGroup = $priceGroupItemCount[$priceGroup] - $selectionCopy[$priceGroup];
if ($remainingInPriceGroup > 0){
$remainingTotal = $total - count($selection);
$priceGroupWeightCopy[$priceGroup] = rand_weighted($remainingInPriceGroup, $remainingTotal);
}
//try to find the solution by recursing
$tryRecursion = $recurse($priceGroupWeightCopy, $selectionCopy, $priceGroupSumCopy);
if ($tryRecursion !== null){
return $tryRecursion;
}
}
}
return null;
};
$selection = $recurse($priceGroupWeight);
if ($selection===null){
echo "there are no possible solutions\n";
} else {
echo "SQL for items:\n";
foreach ($selection as $priceGroup => $numberFromPriceGroup){
$select = "SELECT `items`.*";
$from = "FROM `items`";
$where = "WHERE ROUND(`price`/{$priceGroupIncrement}) = {$priceGroup}";
$orderBy = "ORDER BY rand() ASC";
$limit = "LIMIT {$numberFromPriceGroup}";
$sql = "{$select} {$from} {$where} {$orderBy} {$limit}";
echo "$sql;\n";
}
}
This algorithmic approach should perform much better than a pure SQL Query-based solution, especially once your items table grows.

Answer 2

You'll need to use a HAVING clause -
SELECT item, price, sum(price) as total_price
FROM items
GROUP BY item
HAVING total_price BETWEEN ($value-10) AND ($value+10)
ORDER BY rand() LIMIT 6
Here is an example and here is another example
In this case the sum is always going to be the total of each item (using GROUP BY) which is great if you only have one of each item. If you have more than one of each the sum is going to total all of those items together in the GROUP BY. Based on your original description it is the second query that you're looking for where a customer will be able to see random products within a price range.
It would be best if you provided a table schema (perhaps using SQL Fiddle) and then showed us examples of what you want the results to be.