/// <summary>
/// Iterate over the training data, uniformly sample from users with replacement.
/// TODO
/// </summary>
protected virtual void IterateWithReplacementUniformUser(List <Rating> ratings, MyTable ratingTable,
double gamma = 0.01, double lambda = 0.01, double lambda_bias = 0.01)
{
int[] userIds = ratingTable.GetMainKeyArray().AsParallel().Cast <int>().OrderBy(k => k).ToArray();
int[] itemIds = ratingTable.GetSubKeyArray().AsParallel().Cast <int>().OrderBy(k => k).ToArray();
var random = Core.Random.GetInstance();
int numberOfRatings = ratings.Count;
int userId, itemId, otherItemId;
for (int i = 0; i < numberOfRatings; i++)
{
while (true)
{
// randomly select a user
userId = userIds[random.Next(userIds.Length)];
Hashtable itemsTable = (Hashtable)ratingTable[userId];
// copy
Hashtable itemsTableCopy = new Hashtable(itemsTable); // bugs,2018.02.13
var triple = SampleItemPair(userId, itemsTable, itemIds, random);
itemId = triple.Item2;
otherItemId = triple.Item3;
break;
}
UpdateFactors(userId, itemId, otherItemId, gamma, lambda, lambda_bias);
}
}
/// <summary>
/// Get recommendations based on the trained model?
/// </summary>
/// <param name="ratingTable"></param>
/// <param name="N"></param>
/// <returns></returns>
protected List <Rating> GetRecommendations(MyTable ratingTable, int N = 10)
{
List <Rating> recommendedItems = new List <Rating>();
var itemIds = ratingTable.GetSubKeyArray().AsParallel().Cast <int>().ToArray();
var userIds = ratingTable.GetMainKeyArray().AsParallel().Cast <int>().ToArray();
Parallel.ForEach(userIds, userId =>
{
Hashtable Nu = (Hashtable)ratingTable[userId]; // ratings of user u
List <Rating> predictedRatings = new List <Rating>();
foreach (int itemId in itemIds)
{
if (!Nu.ContainsKey(itemId))
{
double p = Predict(userId, itemId);
predictedRatings.Add(new Rating(userId, itemId, p));
}
}
List <Rating> sortedLi = predictedRatings.OrderByDescending(r => r.Score).ToList();
lock (recommendedItems)
{
recommendedItems.AddRange(sortedLi.GetRange(0, Math.Min(sortedLi.Count, N)));
}
});
return(recommendedItems);
}
/// <summary>
/// Randomly sample negative ratings for each user from his/her rated ratings.
/// Recommender systems in action, p84, vector xiang
/// </summary>
/// <param name="ratings">positive ratings</param>
/// <param name="ratio">ratio = #(negative samples) / #(positive samples)</param>
/// <param name="verbose"></param>
/// <returns></returns>
public static List <Rating> RandomSelectNegativeSamples(List <Rating> ratings, int ratio = 1, bool verbose = false)
{
if (verbose)
{
Console.WriteLine("ratio,{0}", ratio);
}
List <Rating> positiveRatings = new List <Rating>();
foreach (Rating r in ratings)
{
positiveRatings.Add(new Rating(r.UserId, r.ItemId, 1.0));
}
MyTable ratingTable = GetRatingTable(positiveRatings);
int[] items = ratingTable.GetSubKeyArray().AsParallel().Cast <int>().ToArray();
var random = Core.Random.GetInstance();
foreach (int uId in ratingTable.Keys)
{
Hashtable subTable = (Hashtable)ratingTable[uId];
int counter = 0, ratedItems = subTable.Count;
while ((counter < ratedItems * ratio) && (counter < items.Length - ratedItems))
{
int iId = items[random.Next(items.Length)];
if (!subTable.ContainsKey(iId))
{
subTable.Add(iId, 0.0); // negative samples
counter++;
}
}
}
List <Rating> samples = new List <Rating>();
foreach (int uId in ratingTable.Keys)
{
Hashtable subTable = (Hashtable)ratingTable[uId];
foreach (int iId in subTable.Keys)
{
double score = (double)subTable[iId];
samples.Add(new Rating(uId, iId, score));
}
}
return(samples);
}
protected List <Rating> GetRecommendations(MyTable ratingTable, double miu, int N = 10, bool multiThread = false)
{
List <Rating> recommendedItems = new List <Rating>();
int[] subKeys = ratingTable.GetSubKeyArray().AsParallel().Cast <int>().ToArray();
if (multiThread)
{
int[] mainKeys = ratingTable.GetMainKeyArray().AsParallel().Cast <int>().ToArray();
Parallel.ForEach(mainKeys, userId =>
{
Hashtable Nu = (Hashtable)ratingTable[userId]; // ratings of user u
List <Rating> predictedRatings = new List <Rating>();
foreach (int itemId in subKeys)
{
if (!Nu.ContainsKey(itemId))
{
double p = Predict(userId, itemId, miu);
predictedRatings.Add(new Rating(userId, itemId, p));
}
}
List <Rating> sortedLi = predictedRatings.OrderByDescending(r => r.Score).ToList();
lock (recommendedItems)
{
recommendedItems.AddRange(sortedLi.GetRange(0, Math.Min(sortedLi.Count, N)));
}
});
}
else
{
foreach (int userId in ratingTable.Keys)
{
Hashtable Nu = (Hashtable)ratingTable[userId]; // ratings of user u
List <Rating> predictedRatings = new List <Rating>();
foreach (int itemId in subKeys)
{
if (!Nu.ContainsKey(itemId))
{
double p = Predict(userId, itemId, miu);
predictedRatings.Add(new Rating(userId, itemId, p));
}
}
List <Rating> sortedLi = predictedRatings.OrderByDescending(r => r.Score).ToList();
recommendedItems.AddRange(sortedLi.GetRange(0, Math.Min(sortedLi.Count, N)));
}
}
return(recommendedItems);
}
/// <summary>
/// Iterate over the training data, uniformly sample from user-item pairs without replacement.
/// </summary>
protected virtual void IterateWithoutReplacementUniformPair(MyTable ratingTable, double gamma = 0.01, double lambda = 0.01, double lambda_bias = 0.01)
{
int[] itemIds = ratingTable.GetSubKeyArray().AsParallel().Cast <int>().OrderBy(k => k).ToArray();
var random = Core.Random.GetInstance(); // need to be a static member?
foreach (int userId in ratingTable.Keys)
{
Hashtable itemsTable = (Hashtable)ratingTable[userId];
foreach (int itemId in itemsTable.Keys)
{
// 1. sample a negative feedback for each positive feedback
int otherItemId = SampleOtherItemId(userId, itemsTable, itemIds, random);
UpdateFactors(userId, itemId, otherItemId, gamma, lambda, lambda_bias);
}
}
}
/// <summary>
///
/// </summary>
/// <param name="ratingTable"></param>
/// <param name="ratio"></param>
/// <returns>list of triples: user id - item id - other item id</returns>
protected List <Tuple <int, int, int> > SampleTriples(MyTable ratingTable, int ratio = 100)
{
int[] userIds = ratingTable.GetMainKeyArray().AsParallel().Cast <int>().OrderBy(k => k).ToArray();
int[] itemIds = ratingTable.GetSubKeyArray().AsParallel().Cast <int>().OrderBy(k => k).ToArray();
List <Tuple <int, int, int> > list = new List <Tuple <int, int, int> >();
var random = Core.Random.GetInstance(); // need to be a static member?
for (int i = 0; i < userIds.Length * ratio; i++)
{
// randomly select a user
int userId = userIds[random.Next(userIds.Length)];
Hashtable itemsTable = (Hashtable)ratingTable[userId];
var triple = SampleItemPair(userId, itemsTable, itemIds, random);
list.Add(triple);
}
return(list);
}
/// <summary>
/// Iterate over the training data, uniformly sample from user-item pairs with replacement.
/// </summary>
protected virtual void IterateWithReplacementUniformPair(List <Rating> ratings, MyTable ratingTable, double gamma = 0.01, double lambda = 0.01, double lambda_bias = 0.01)
{
int[] itemIds = ratingTable.GetSubKeyArray().AsParallel().Cast <int>().OrderBy(k => k).ToArray();
var random = Core.Random.GetInstance();
int numberOfRatings = ratings.Count;
for (int ii = 0; ii < numberOfRatings; ii++)
{
Rating r = ratings[random.Next(numberOfRatings)];
int userId = r.UserId;
int itemId = r.ItemId;
Hashtable itemsTable = (Hashtable)ratingTable[userId];
// 1. sample a negative feedback for each positive feedback
int otherItemId = SampleOtherItemId(userId, itemsTable, itemIds, random);
UpdateFactors(userId, itemId, otherItemId, gamma, lambda, lambda_bias);
}
}
