Kadane's algorithm to find subarray with the m

问题:

This question already has an answer here:

• Maximum sum sublist? 12 answers

I have the following implementation of Kadane's algorithm to solve the problem of the maximum subarray of an array:

public static decimal FindBestSubsequence
    (this IEnumerable<decimal> source, out int startIndex, out int endIndex)
{
    decimal result = decimal.MinValue;
    decimal sum = 0;
    int tempStart = 0;

    List<decimal> tempList = new List<decimal>(source);

    startIndex = 0;
    endIndex = 0;

    for (int index = 0; index < tempList.Count; index++)
    {
        sum += tempList[index];
        if ((sum > result) || 
            (sum == result && (endIndex - startIndex) < (index - tempStart)))
        {
            result = sum;
            startIndex = tempStart;
            endIndex = index;
        }
        else if (sum < 0)
        {
            sum = 0;
            tempStart = index + 1;
        }
    }

    return result;
}

It fails when I use a sequence that starts with a negative number like -1, 2, 3 giving a result of 4, [0,2] instead of 5, [1,2].

For the life of me that I cannot find where the error is. Maybe its a defect on the algorithm design?

Thanks in advance.

回答1:

Your initial implementation suffered from unnecessarily complicated and partially wrong checks within the main scan cycle. These checks are two:

• if greater intermediate sum is found, store it constituents as a temporary result;
• independently, if sum got negative, reset it to 0 and prepare to build a new sequence from next scan position.

Refactored FindBestSubsequence method implementation is listed below:

public static decimal FindBestSubsequence (this IEnumerable<decimal> source, out int startIndex, out int endIndex)
{
    decimal result = decimal.MinValue;
    decimal sum = 0;
    int tempStart = 0;

    List<decimal> tempList = new List<decimal>(source);

    startIndex = 0;
    endIndex = 0;

    for (int index = 0; index < tempList.Count; index++)
    {
        sum += tempList[index];
        if (sum > result)
        {
            result = sum;
            startIndex = tempStart;
            endIndex = index;
        }
        if (sum < 0)
        {
            sum = 0;
            tempStart = index + 1;
        }
    }

    return result;
}

Now not only for -1,2,3 the code above produces correct answer 5,[1,2] but also it correctly processes arrays of all negative numbers without any extra code: entering -10,-2,-3 will return -2,[1,1].

回答2:

In your example you always have sum > result even if sum<0 in the first iteration of The loop because 0 > decimal.MinValue.

So you never go to your second case.-

You need to change the first if by adding a condition sum > 0:

if ((sum >0 ) & ((sum > result) || 
    (sum == result && (endIndex - startIndex) < (index - tempStart))))
{
    ...
}
else if (sum < 0)
{
    ...
}

Update:

As explained in my comment you can just change the initialisation of result to 0 :

decimal result = 0;

From wikipedia :

This subarray is either empty (in which case its sum is zero) or consists of one more element than the maximum subarray ending at the previous position

Therefore if the array contains only negative numbers the solution is an empty subarray with sum 0.

回答3:

Change this line:

decimal result = decimal.MinValue;

to

decimal result = 0;

回答4:

For each position you should take the maximum of the value there (from the original sequence) and your sum as you have written it. If the original number is bigger, then it's better to start summing 'from beginning', i.e. sum = max(sum+tempList[index],tempList[index]); Then you won't need the case for sum < 0 at all.

回答5:

At the end this is how I corrected the algorithm to handle all the scenarios, just in case it helps to someone:

    public static decimal FindBestSubsequence (this IEnumerable<decimal> source, out int startIndex, out int endIndex)
    {
        decimal result = decimal.MinValue;
        decimal sum = 0;
        int tempStart = 0;

        List<decimal> tempList = new List<decimal>(source);

        if (tempList.TrueForAll(v => v <= 0))
        {
            result = tempList.Max();
            startIndex = endIndex = tempList.IndexOf(result);
        }
        else
        {
            startIndex = 0;
            endIndex = 0;

            for (int index = 0; index < tempList.Count; index++)
            {
                sum += tempList[index];

                if (sum > 0 && sum > result || (sum == result && (endIndex - startIndex) < (index - tempStart)))
                {
                    result = sum;
                    startIndex = tempStart;
                    endIndex = index;
                }
                else if (sum < 0)
                {
                    sum = 0;
                    tempStart = index + 1;
                }
            }
        }

        return result;
    }

回答6:

Built upon Gene Belitski's answer and comments:

    public static void Main()
    {
        var seq = new[] { -10M, -2M, -3M };
        var stuff = seq.FindBestSubsequence();

        Console.WriteLine(stuff.Item1 + " " + stuff.Item2 + " " + stuff.Item3);
        Console.ReadLine();
    }

    public static Tuple<decimal, long, long> FindBestSubsequence(this IEnumerable<decimal> source)
    {
        var result = new Tuple<decimal, long, long>(decimal.MinValue, -1L, -1L);

        if (source == null)
        {
            return result;
        }

        var sum = 0M;
        var tempStart = 0L;
        var index = 0L;

        foreach (var item in source)
        {
            sum += item;
            if (sum > result.Item1)
            {
                result = new Tuple<decimal, long, long>(sum, tempStart, index);
            }

            if (sum < 0)
            {
                sum = 0;
                tempStart = index + 1;
            }

            index++;
        }

        return result;
    }