/// <summary>
/// Add a data object with associated category and value.
/// </summary>
/// <param name="data">The data object which will be returned from GetBestMatch</param>
/// <param name="category">The category associated with this value</param>
/// <param name="value">The value</param>
public void Add(T data, string category, decimal value)
{
CategorySpread spread = null;
if (!categories.TryGetValue(category, out spread))
{
spread = new CategorySpread();
categories[category] = spread;
}
spread.Miniumum = Math.Min(spread.Miniumum, value);
spread.Maximum = Math.Max(spread.Maximum, value);
spread.Count++;
spread.Total += value;
spread.Values.Add((double)value);
spread.Data.Add(data);
total++;
}
/// <summary>
/// Get the data object that has the most likely category for the given value.
/// </summary>
/// <param name="value">The value we want to match</param>
/// <returns>The data object</returns>
public T GetBestMatch(decimal value)
{
if (categories.Count == 0)
{
return(default(T));
}
CategorySpread result = null;
if (categories.Count > 1)
{
int max = 0;
foreach (CategorySpread spread in categories.Values)
{
max = Math.Max(max, spread.Values.Count);
}
// trim out any set that is less than 1/5th the size of the maximum set.
// those are just too rare to be useful as a best match.
int min = max / 5;
decimal smallestDeviation = decimal.MaxValue;
foreach (CategorySpread spread in categories.Values.Where(s => s.Count > min))
{
decimal mean = spread.Mean;
decimal deviation = spread.StandardDeviation;
if (deviation == 0)
{
// can't use it, not enough samples.
continue;
}
// how many standard deviations is the given value from the mean?
decimal deviations = Math.Abs(Math.Abs(value) - Math.Abs(mean)) / deviation;
// Pick the category where the given value is the closest to one standard deviation
// from the mean. Dividing by the standard deviation means if we have hundreds of samples
// in one category with a broad standard deviation, like category "Food:Groceries", and the
// value falls close to the mean then this category is favored over something that has only
// a few samples and a narrow standard deviation.
if (result == null || deviations < smallestDeviation)
{
smallestDeviation = deviations;
result = spread;
}
}
}
if (result == null)
{
// return something if we can (might have just one category).
return(categories.Values.LastOrDefault().Data.LastOrDefault());
}
if (result != null)
{
return(result.Data.LastOrDefault());
}
return(default(T));
}
