private SingleUserEvaluationResults GetSingleUserEvaluationResults <TInput, TRecommendation>(
IRecommendationSource <TInput, IEnumerable <TRecommendation>, TRecommendation> recSource,
TInput user,
IUserInputClassifier <TInput> goodBadClassifier,
Func <ClassifiedUserInput <TInput>, double, ItemsForInputAndEvaluation <TInput> > inputDivisionFunc,
int numRecsToTryToGet)
where TInput : IInputForUser
where TRecommendation : IRecommendation
{
// Divide into liked and unliked.
// Set aside a random X% of the liked and a random X% of the unliked
ClassifiedUserInput <TInput> classified = goodBadClassifier.Classify(user);
ItemsForInputAndEvaluation <TInput> divided = inputDivisionFunc(classified, FractionOfInputToSetAsideForEvaluation);
// Get top N recommendations
// Keep count of hits and false positives
List <int> recommendedIds = new List <int>();
int truePositivesForThisUser = 0;
int falsePositivesForThisUser = 0;
int unknownsForThisUser = 0;
foreach (TRecommendation recommendation in recSource.GetRecommendations(divided.ItemsForInput, numRecsToTryToGet))
{
recommendedIds.Add(recommendation.ItemId);
if (divided.LikedItemsForEvaluation.Contains(recommendation.ItemId))
{
truePositivesForThisUser++;
}
else if (divided.UnlikedItemsForEvaluation.Contains(recommendation.ItemId))
{
falsePositivesForThisUser++;
}
else
{
unknownsForThisUser++;
}
}
int falseNegativesForThisUser = divided.LikedItemsForEvaluation.Count - truePositivesForThisUser;
SingleUserEvaluationResults results = new SingleUserEvaluationResults()
{
TruePositives = truePositivesForThisUser,
FalsePositives = falsePositivesForThisUser,
Unknowns = unknownsForThisUser,
FalseNegatives = falseNegativesForThisUser,
};
return(results);
}
public IPositiveFeedbackForUser AsPositiveFeedback(IUserInputClassifier <MalUserListEntries> classifier, Predicate <int> additionalOkToRecommendPredicate)
{
ClassifiedUserInput <MalUserListEntries> classified = classifier.Classify(this);
HashSet <int> basicFeedback = new HashSet <int>(classified.Liked.Entries.Select(itemIdEntryPair => itemIdEntryPair.Key));
if (additionalOkToRecommendPredicate == null)
{
return(new BasicPositiveFeedbackForUserWithOkToRecommendPredicate(basicFeedback, ItemIsOkToRecommend));
}
else
{
return(new BasicPositiveFeedbackForUserWithOkToRecommendPredicate(basicFeedback, (itemId) =>
ItemIsOkToRecommend(itemId) && additionalOkToRecommendPredicate(itemId)));
}
}
public static ItemsForInputAndEvaluation <MalUserListEntries> DivideClassifiedForInputAndEvaluation(
ClassifiedUserInput <MalUserListEntries> classifiedInput, double fractionToSetAsideForEvaluation)
{
IDictionary <int, MalListEntry> entriesForInput = new Dictionary <int, MalListEntry>();
HashSet <int> likedAnimesForEvaluation = new HashSet <int>();
HashSet <int> unlikedAnimesForEvaluation = new HashSet <int>();
// Recommender could potentially use someone's "plan to watch" list to infer information about the user's taste...or something.
foreach (KeyValuePair <int, MalListEntry> otherEntry in classifiedInput.Other.Entries)
{
entriesForInput.Add(otherEntry);
}
List <int> likedAnimeIds = new List <int>(classifiedInput.Liked.Entries.Keys);
List <int> unlikedAnimeIds = new List <int>(classifiedInput.NotLiked.Entries.Keys);
likedAnimeIds.Shuffle();
unlikedAnimeIds.Shuffle();
int numLikedForEvaluation = (int)(likedAnimeIds.Count * fractionToSetAsideForEvaluation);
int numUnlikedForEvaluation = (int)(unlikedAnimeIds.Count * fractionToSetAsideForEvaluation);
for (int i = 0; i < numLikedForEvaluation; i++)
{
likedAnimesForEvaluation.Add(likedAnimeIds[i]);
}
for (int i = numLikedForEvaluation; i < likedAnimeIds.Count; i++)
{
entriesForInput[likedAnimeIds[i]] = classifiedInput.Liked.Entries[likedAnimeIds[i]];
}
for (int i = 0; i < numUnlikedForEvaluation; i++)
{
unlikedAnimesForEvaluation.Add(unlikedAnimeIds[i]);
}
for (int i = numUnlikedForEvaluation; i < unlikedAnimeIds.Count; i++)
{
entriesForInput[unlikedAnimeIds[i]] = classifiedInput.NotLiked.Entries[unlikedAnimeIds[i]];
}
return(new ItemsForInputAndEvaluation <MalUserListEntries>()
{
ItemsForInput = new MalUserListEntries(entriesForInput, classifiedInput.Liked.AnimesEligibleForRecommendation,
classifiedInput.Liked.MalUsername, classifiedInput.Liked.OkToRecommendPredicate),
LikedItemsForEvaluation = likedAnimesForEvaluation,
UnlikedItemsForEvaluation = unlikedAnimesForEvaluation
});
}
public static ItemsForInputAndEvaluation<MalUserListEntries> DivideClassifiedForInputAndEvaluation(
ClassifiedUserInput<MalUserListEntries> classifiedInput, double fractionToSetAsideForEvaluation)
{
IDictionary<int, MalListEntry> entriesForInput = new Dictionary<int, MalListEntry>();
HashSet<int> likedAnimesForEvaluation = new HashSet<int>();
HashSet<int> unlikedAnimesForEvaluation = new HashSet<int>();
// Recommender could potentially use someone's "plan to watch" list to infer information about the user's taste...or something.
foreach (KeyValuePair<int, MalListEntry> otherEntry in classifiedInput.Other.Entries)
{
entriesForInput.Add(otherEntry);
}
List<int> likedAnimeIds = new List<int>(classifiedInput.Liked.Entries.Keys);
List<int> unlikedAnimeIds = new List<int>(classifiedInput.NotLiked.Entries.Keys);
likedAnimeIds.Shuffle();
unlikedAnimeIds.Shuffle();
int numLikedForEvaluation = (int)(likedAnimeIds.Count * fractionToSetAsideForEvaluation);
int numUnlikedForEvaluation = (int)(unlikedAnimeIds.Count * fractionToSetAsideForEvaluation);
for (int i = 0; i < numLikedForEvaluation; i++)
{
likedAnimesForEvaluation.Add(likedAnimeIds[i]);
}
for (int i = numLikedForEvaluation; i < likedAnimeIds.Count; i++)
{
entriesForInput[likedAnimeIds[i]] = classifiedInput.Liked.Entries[likedAnimeIds[i]];
}
for (int i = 0; i < numUnlikedForEvaluation; i++)
{
unlikedAnimesForEvaluation.Add(unlikedAnimeIds[i]);
}
for (int i = numUnlikedForEvaluation; i < unlikedAnimeIds.Count; i++)
{
entriesForInput[unlikedAnimeIds[i]] = classifiedInput.NotLiked.Entries[unlikedAnimeIds[i]];
}
return new ItemsForInputAndEvaluation<MalUserListEntries>()
{
ItemsForInput = new MalUserListEntries(entriesForInput, classifiedInput.Liked.AnimesEligibleForRecommendation,
classifiedInput.Liked.MalUsername, classifiedInput.Liked.OkToRecommendPredicate),
LikedItemsForEvaluation = likedAnimesForEvaluation,
UnlikedItemsForEvaluation = unlikedAnimesForEvaluation
};
}
