/// <summary>
/// Creates a new hierarchical N-Gram.
/// </summary>
/// <param name="nValue">The N in N-Gram (window size + 1).</param>
/// <param name="threshold">
/// Minimum number of observations for a specific N-Gram to return a
/// prediction.
/// </param>
public HierarchNGram(int nValue, int threshold)
{
// Keep the N value
NValue = nValue;
// Keep the threshold
this.threshold = threshold;
// Instantiate the array of N-Grams
predictors = new NGram <T> [nValue];
// Instantiate the individual N-Grams
for (int i = 0; i < nValue; i++)
{
predictors[i] = new NGram <T>(i + 1);
}
}
/// <summary>
/// Get the most likely action given a sequence of actions.
/// </summary>
/// <param name="actions">
/// The actions list, which should be at least of size 1.
/// </param>
/// <returns>
/// The most likely action for the given a sequence of actions.
/// </returns>
public T GetMostLikely(IReadOnlyList <T> actions)
{
// Declare variable for best action and set it to its default value
T bestAction = default;
// Go through the various Ni-Grams
for (int i = 0; i < NValue; i++)
{
// Are there enough actions for the current Ni-Gram?
if (actions.Count >= NValue - i - 1)
{
// Get current Ni-Gram
NGram <T> p = predictors[NValue - i - 1];
// Create a view containing only the actions for the
// current Ni-Gram
ReadOnlyListSegment <T> subactions =
new ReadOnlyListSegment <T>(
actions,
actions.Count + i + 1 - NValue,
NValue - i - 1);
// Get frequency of action sequence in the current Ni-Gram
int numActions = p.GetActionsFrequency(subactions);
// Is that frequency larger than the threshold?
if (numActions > threshold)
{
// Then use this action
bestAction = p.GetMostLikely(subactions);
break;
}
}
}
// Return the best action
return(bestAction);
}
