///
protected virtual void InitModel()
{
user_factors = new Matrix<float>(MaxUserID + 1, NumFactors);
item_factors = new Matrix<float>(MaxItemID + 1, NumFactors);
user_factors.InitNormal(InitMean, InitStdDev);
item_factors.InitNormal(InitMean, InitStdDev);
}
///
protected virtual void InitModel()
{
item_weights = new Matrix<float>(MaxItemID + 1, MaxItemID + 1);
item_weights.InitNormal(InitMean, InitStdDev);
// set diagonal elements to 0
for(int i = 0; i <= MaxItemID; i++)
{
item_weights[i, i] = 0;
}
}
///
protected internal override void InitModel()
{
base.InitModel();
x = new Matrix<float>(item_attributes.NumberOfColumns, NumFactors);
x.InitNormal(InitMean, InitStdDev);
q = new Matrix<float>(MaxItemID + 1, NumFactors);
q.InitNormal(InitMean, InitStdDev);
// set factors to zero for items without training examples
for (int i = 0; i < ratings.CountByItem.Count; i++)
if (ratings.CountByItem[i] == 0)
q.SetRowToOneValue(i, 0);
}
///
protected override void InitModel()
{
base.InitModel ();
user_factors = null;
item_factors = null;
p = new Matrix<float> (MaxUserID + 1, NumFactors);
p.InitNormal (InitMean, InitStdDev);
y = new Matrix<float> (MaxItemID + 1, NumFactors);
y.InitNormal (InitMean, InitStdDev);
x = new Matrix<float> (item_attributes.NumberOfColumns, NumFactors);
x.InitNormal (InitMean, InitStdDev);
q = new Matrix<float> (MaxItemID + 1, NumFactors);
q.InitNormal (InitMean, InitStdDev);
int num_attributes = item_attributes.NumberOfColumns;
x_reg = new float[num_attributes];
for (int attribute_id = 0; attribute_id < num_attributes; attribute_id++)
x_reg [attribute_id] = FrequencyRegularization? (RegX / (float)(1 + Math.Exp(-0.005*item_attributes.NumEntriesByColumn (attribute_id)))) : RegX;
y_reg = new float[MaxItemID + 1];
for (int item_id = 0; item_id <= MaxItemID; item_id++) {
var feedback_count_by_item = Feedback.ItemMatrix [item_id];
if (feedback_count_by_item.Count > 0)
y_reg [item_id] = FrequencyRegularization ? (float)(RegY / Math.Sqrt (feedback_count_by_item.Count)) : RegY;
else
y_reg [item_id] = 0;
}
Console.Write("Learning attributes...");
BPRLinear learnAttr = new BPRLinear();
learnAttr.Feedback = Feedback;
learnAttr.ItemAttributes = item_attributes;
learnAttr.NumIter = NumIter;//10;
learnAttr.LearnRate = LearnRate;//0.05f;
learnAttr.Regularization = 0.015f;//0.001f;
learnAttr.Train();
item_attribute_weight_by_user = learnAttr.ItemAttributeWeights;
learnAttr = null;
Console.WriteLine ("Done");
}
///
protected internal override void InitModel()
{
base.InitModel();
p = new Matrix<float>(MaxUserID + 1, NumFactors);
p.InitNormal(InitMean, InitStdDev);
y = new Matrix<float>(MaxItemID + 1, NumFactors);
y.InitNormal(InitMean, InitStdDev);
// set factors to zero for items without training examples
for (int i = 0; i < ratings.CountByItem.Count; i++)
if (ratings.CountByItem[i] == 0)
y.SetRowToOneValue(i, 0);
for (int i = ratings.CountByItem.Count; i <= MaxItemID; i++)
{
y.SetRowToOneValue(i, 0);
item_factors.SetRowToOneValue(i, 0);
}
// set factors to zero for users without training examples (rest is done in MatrixFactorization.cs)
for (int u = ratings.CountByUser.Count; u <= MaxUserID; u++)
p.SetRowToOneValue(u, 0);
user_bias = new float[MaxUserID + 1];
item_bias = new float[MaxItemID + 1];
}
///
protected override void InitModel()
{
x = new Matrix<float>(MaxUserID + 1, NumFactors);
x.InitNormal(InitMean, InitStdDev);
// set factors to zero for users without training examples
for (int user_id = 0; user_id < x.NumberOfRows; user_id++)
if (user_id > ratings.MaxUserID || ratings.CountByUser[user_id] == 0)
x.SetRowToOneValue(user_id, 0);
base.InitModel();
}
///
protected internal override void InitModel()
{
y = new Matrix<float>(MaxItemID + 1, NumFactors);
y.InitNormal(InitMean, InitStdDev);
// set factors to zero for items without training examples
for (int item_id = 0; item_id < y.NumberOfRows; item_id++)
if (item_id > ratings.MaxItemID || ratings.CountByItem[item_id] == 0)
y.SetRowToOneValue(item_id, 0);
base.InitModel();
}
/// <summary>Initialize the model data structure</summary>
protected virtual void InitModel()
{
// init factor matrices
user_factors = new Matrix<float>(MaxUserID + 1, NumFactors);
item_factors = new Matrix<float>(MaxItemID + 1, NumFactors);
user_factors.InitNormal(InitMean, InitStdDev);
item_factors.InitNormal(InitMean, InitStdDev);
// set factors to zero for users and items without training examples
for (int u = 0; u < ratings.CountByUser.Count; u++)
if (ratings.CountByUser[u] == 0)
user_factors.SetRowToOneValue(u, 0);
for (int i = 0; i < ratings.CountByItem.Count; i++)
if (ratings.CountByItem[i] == 0)
item_factors.SetRowToOneValue(i, 0);
}
void LearnUserAttributeToFactorMapping()
{
// no mapping of no user attributes present
if (user_attributes.NumberOfEntries == 0)
return;
// create attribute-to-factor weight matrix
this.user_attribute_to_factor = new Matrix<float>(NumUserAttributes + 1, num_factors);
Console.Error.WriteLine("num_user_attributes=" + NumUserAttributes);
// store the results of the different runs in the following array
var old_user_attribute_to_factor = new Matrix<float>[num_init_mapping];
Console.Error.WriteLine("Will use {0} examples ...", num_iter_mapping * MaxUserID);
var old_rmse_per_factor = new double[num_init_mapping][];
for (int h = 0; h < num_init_mapping; h++)
{
user_attribute_to_factor.InitNormal(InitMean, InitStdDev);
Console.Error.WriteLine("----");
for (int i = 0; i < num_iter_mapping * MaxUserID; i++)
UpdateUserMapping();
ComputeUserMappingFit();
old_user_attribute_to_factor[h] = new Matrix<float>(user_attribute_to_factor);
old_rmse_per_factor[h] = ComputeUserMappingFit();
}
var min_rmse_per_factor = new double[num_factors];
for (int i = 0; i < num_factors; i++)
min_rmse_per_factor[i] = double.MaxValue;
var best_factor_init = new int[num_factors];
// find best factor mappings:
for (int i = 0; i < num_init_mapping; i++)
{
for (int j = 0; j < num_factors; j++)
{
if (old_rmse_per_factor[i][j] < min_rmse_per_factor[j])
{
min_rmse_per_factor[j] = old_rmse_per_factor[i][j];
best_factor_init[j] = i;
}
}
}
// set the best weight combinations for each factor mapping
for (int i = 0; i < num_factors; i++)
{
Console.Error.WriteLine("Factor {0}, pick {1}", i, best_factor_init[i]);
user_attribute_to_factor.SetColumn(i, old_user_attribute_to_factor[best_factor_init[i]].GetColumn(i));
}
Console.Error.WriteLine("----");
ComputeUserMappingFit();
ComputeFactorsForNewUsers();
}
///
protected internal override void InitModel()
{
base.InitModel ();
p = new Matrix<float> (MaxUserID + 1, NumFactors);
p.InitNormal (InitMean, InitStdDev);
y = new Matrix<float> (MaxItemID + 1, NumFactors);
y.InitNormal (InitMean, InitStdDev);
// set factors to zero for items without training examples
for (int i = 0; i < ratings.CountByItem.Count; i++)
if (ratings.CountByItem [i] == 0)
y.SetRowToOneValue (i, 0);
for (int i = ratings.CountByItem.Count; i <= MaxItemID; i++) {
y.SetRowToOneValue (i, 0);
item_factors.SetRowToOneValue (i, 0);
}
// set factors to zero for users without training examples (rest is done in MatrixFactorization.cs)
for (int u = ratings.CountByUser.Count; u <= MaxUserID; u++) {
p.SetRowToOneValue (u, 0);
}
user_bias = new float[MaxUserID + 1];
item_bias = new float[MaxItemID + 1];
h = new Matrix<float>[AdditionalUserAttributes.Count + 1];
h [0] = new Matrix<float> (UserAttributes.NumberOfColumns, ItemAttributes.NumberOfColumns);
h [0].InitNormal (InitMean, InitStdDev);
for (int d = 0; d < AdditionalUserAttributes.Count; d++) {
h [d + 1] = new Matrix<float> (AdditionalUserAttributes [d].NumberOfColumns, ItemAttributes.NumberOfColumns);
h [d + 1].InitNormal (InitMean, InitStdDev);
}
}
///
protected override void InitModel()
{
base.InitModel ();
user_factors = null;
item_factors = null;
p = new Matrix<float> (MaxUserID + 1, NumFactors);
p.InitNormal (InitMean, InitStdDev);
q = new Matrix<float> (MaxItemID + 1, NumFactors);
q.InitNormal (InitMean, InitStdDev);
Console.Write("Learning attributes...");
BPRLinear learnAttr = new BPRLinear();
learnAttr.Feedback = Feedback;
learnAttr.ItemAttributes = item_attributes;
learnAttr.NumIter = NumIter;//10;
learnAttr.LearnRate = LearnRate;//0.05f;
learnAttr.Regularization = 0.015f;//0.001f;
learnAttr.Train();
item_attribute_weight_by_user = learnAttr.ItemAttributeWeights;
learnAttr = null;
Console.WriteLine ("Done");
}
///
protected override void InitModel()
{
base.InitModel();
p = new Matrix<float>(MaxUserID + 1, NumFactors);
p.InitNormal(InitMean, InitStdDev);
y = new Matrix<float>(MaxItemID + 1, NumFactors);
y.InitNormal(InitMean, InitStdDev);
// set factors to zero for items without training examples
for (int i = 0; i <= MaxItemID; i++)
if (ratings.CountByItem[i] == 0)
y.SetRowToOneValue(i, 0);
user_bias = new float[MaxUserID + 1];
item_bias = new float[MaxItemID + 1];
}
