public static List<Sample> Resampling (Segment segment)
{
List<Sample> result = new List<Sample>();
Matrix[] Evaluation_Matrix = new Matrix[10];
Matrix[] Training_Segment_Matrix = new Matrix[10];
Matrix[] Training_Root_Matrix = new Matrix[10];
for (int i = 0; i < 10; i++)
{
Evaluation_Matrix[i] = new Matrix(segment.data.rows, segment.data.cols);
Training_Segment_Matrix[i] = segment.data.Duplicate();
Training_Root_Matrix[i] = segment.data.Duplicate();
}
int count_segment = 0;
int need = segment.data.CountCells() / 10;
Random rand = new Random();
for (int i = 0; i < segment.data.rows; i++)
{
for (int j = 0; j < segment.data.cols; j++)
{
if (segment.data[i, j] != 0)
{
bool isUsed = false;
while (!isUsed)
{
int k = rand.Next(0, 10);
if (Evaluation_Matrix[k].CountCells() < need + 1)
{
isUsed = true;
Evaluation_Matrix[k][i, j] = segment.data[i, j];
Training_Segment_Matrix[k][i, j] = 0;
Training_Root_Matrix[k][i, j] = 0;
}
}
}
}
}
for (int i = 0; i < 10; i++)
{
Segment Evaluation_Segment = new Segment(segment, Evaluation_Matrix[i]);
Segment Training_Segment = new Segment(segment, Training_Segment_Matrix[i]);
Sample sample = new Sample(Evaluation_Segment, Training_Segment);
sample.Training_root = new Segment(segment, Training_Root_Matrix[i]);
result.Add(sample);
}
return result;
}
public double MSE(Segment segment)
{
double sum = 0;
for (int u = 0; u < Uij.Length; u++)
{
sum += Math.Pow(segment.data[Uij[u], j] - ExpertScore(segment.data[Uij[u], j], segment), 2);
}
//sum = 0
if (sum == 0)
return 0.01;
else
return sum / Uij.Length;
}
public Sample(DataTable EvalTable, DataTable TrainingTable)
{
Segment EvalSegment = new Segment();
EvalSegment.data = new Matrix(EvalTable.Rows.Count - 1, EvalTable.Columns.Count - 2);
EvalSegment.user_id = new int[EvalTable.Rows.Count - 1];
EvalSegment.item_id = new string[EvalTable.Columns.Count - 2];
EvalSegment.avgRatingByItem = new double[EvalTable.Columns.Count - 2];
EvalSegment.avgRatingByUser = new double[EvalTable.Rows.Count - 1];
//item
for (int i = 0; i < EvalSegment.item_id.Length; i++)
{
EvalSegment.item_id[i] = EvalTable.Columns[i + 2].ColumnName.Trim();
EvalSegment.avgRatingByItem[i] = Convert.ToDouble(EvalTable.Rows[0][i + 2] == "" ? 0 : EvalTable.Rows[0][i + 2]);
}
for (int i = 0; i < EvalSegment.data.rows; i++)
{
EvalSegment.user_id[i] = Convert.ToInt32(EvalTable.Rows[i + 1][0]);
EvalSegment.avgRatingByUser[i] = Convert.ToDouble(EvalTable.Rows[i + 1][1] == "" ? 0 : EvalTable.Rows[i + 1][1]);
for (int j = 0; j < EvalSegment.data.cols; j++)
EvalSegment.data[i, j] = Convert.ToDouble(EvalTable.Rows[i + 1][j + 2] == "" ? 0 : EvalTable.Rows[i + 1][j + 2]);
}
Segment TrainingSegment = new Segment();
TrainingSegment.data = new Matrix(TrainingTable.Rows.Count - 1, TrainingTable.Columns.Count - 2);
TrainingSegment.user_id = new int[TrainingTable.Rows.Count - 1];
TrainingSegment.item_id = new string[TrainingTable.Columns.Count - 2];
TrainingSegment.avgRatingByItem = new double[TrainingTable.Columns.Count - 2];
TrainingSegment.avgRatingByUser = new double[TrainingTable.Rows.Count - 1];
//item
for (int i = 0; i < TrainingSegment.item_id.Length; i++)
{
TrainingSegment.item_id[i] = TrainingTable.Columns[i + 2].ColumnName.Trim();
TrainingSegment.avgRatingByItem[i] = Convert.ToDouble(TrainingTable.Rows[0][i + 2] == "" ? 0 : TrainingTable.Rows[0][i + 2]);
}
for (int i = 0; i < TrainingSegment.data.rows; i++)
{
TrainingSegment.user_id[i] = Convert.ToInt32(TrainingTable.Rows[i + 1][0]);
TrainingSegment.avgRatingByUser[i] = Convert.ToDouble(TrainingTable.Rows[i + 1][1] == "" ? 0 : TrainingTable.Rows[i + 1][1]);
for (int j = 0; j < TrainingSegment.data.cols; j++)
TrainingSegment.data[i, j] = Convert.ToDouble(TrainingTable.Rows[i + 1][j + 2] == "" ? 0 : TrainingTable.Rows[i + 1][j + 2]);
}
this.Evaluation = EvalSegment;
this.Training = TrainingSegment;
}
public double ExpertScore(double r_aj, Segment segment)// i: item cần tính, j: item mà active user đã rate với giá trị x
{
double alpha = 0;
double beta = 0;
double formular_alpha_1 = 0, formular_alpha_2 = 0;
for (int u = 0; u < Uij.Length; u++)
{
formular_alpha_1 += (segment.data[Uij[u], i] - segment.avgRatingByItem[i]) * (segment.data[Uij[u], j] - segment.avgRatingByItem[j]);
formular_alpha_2 += Math.Pow(segment.data[Uij[u], i] - segment.avgRatingByItem[i], 2);
}
alpha = formular_alpha_1 / formular_alpha_2;
beta = segment.avgRatingByItem[i] - alpha * segment.avgRatingByItem[j];
return alpha * r_aj + beta;
}
public Segment (Segment segment, Matrix data)
{
this.id = segment.id;
this.time = segment.time;
this.budget = segment.budget;
this.companion = segment.companion;
//this.familiarity = segment.familiarity;
//this.mood = segment.mood;
//this.travelLength = segment.travelLength;
this.weather = segment.weather;
this.user_id = segment.user_id;
this.item_id = segment.item_id;
this.avgRatingByItem = segment.avgRatingByItem;
this.avgRatingByUser = segment.avgRatingByUser;
this.data = data;
}
public double ExpertScore(Segment segment)// i: item cần tính, j: item expert
{
double alpha = 0;
double beta = 0;
double formular_alpha_1 = 0, formular_alpha_2 = 0;
double r_uj = 0;//trung bình rating cho item j của các user trên tập giao của i, j
for (int u = 0; u < Uij.Length; u++)
{
formular_alpha_1 += (segment.data[u, i] - segment.avgRatingByItem[i]) * (segment.data[u, j] - segment.avgRatingByItem[j]);
formular_alpha_2 += Math.Pow(segment.data[u, i] - segment.avgRatingByItem[i], 2);
r_uj += segment.data[u, j];
}
r_uj = r_uj / Uij.Length;
alpha = formular_alpha_1 / formular_alpha_2;
beta = segment.avgRatingByItem[i] - alpha * segment.avgRatingByItem[j];
return alpha * r_uj + beta;
}
int[] Uij {get; set;} //u_index (not id)
public Regression(int i, int j, Segment segment)
{
this.i = i;
this.j = j;
/*int i_index = -1, j_index = -1;
for (int index = 0; index < segment.item_id.Length; index++)
{
if (Convert.ToInt32(segment.item_id[index].Trim()) == i)
i_index = index;
if (Convert.ToInt32(segment.item_id[index].Trim()) == j)
j_index = index;
if (i_index != -1 && j_index != -1)
break;
}*/
List<int> L_Uij = new List<int>();
for (int index = 0; index < segment.data.rows; index++)
{
//if (segment.data[index, i_index] != 0 && segment.data[index, j_index] != 0)
if (segment.data[index, i] != 0 && segment.data[index, j] != 0)
L_Uij.Add(index);
}
this.Uij = L_Uij.ToArray();
}
public Sample(Segment Eval, Segment Training)
{
this.Evaluation = Eval;
this.Training = Training;
}
public static double Correlation(int i, int j, int k, Segment segment)
{
Regression R_ij = new Regression(i, j, segment);
Regression R_ik = new Regression(i, k, segment);
Regression R_jk = new Regression(j, k, segment);
double formular1 = 0, formular2 = 0;
double Cijj = R_ij.MSE(segment);
double Cikk = R_ik.MSE(segment);
double temp = 0; //EU_jk[(F_ji - F_ki)^2]
int count_temp = 0;
for (int u = 0; u < R_jk.Uij.Length; u++)
{
int u_index = R_jk.Uij[u];
if (segment.data[u_index, i] != 0)
{
temp += Math.Pow(R_ij.ExpertScore(segment.data[u_index, i], segment) - R_ik.ExpertScore(segment.data[u_index, i], segment), 2);
count_temp++;
}
}
//temp = temp / R_jk.Uij.Length;
temp = temp / count_temp;
formular1 = Cijj + Cikk - temp;
formular2 = 2 * Math.Sqrt(Cijj * Cikk);
return formular1 / formular2;
}
public static double Prediction(int active_user_id, int item_id, Segment segment)
{
//index of item_id in segment data
int item_id_index = -1;
for (int i = 0; i < segment.item_id.Length; i++)
{
if (Convert.ToInt32(segment.item_id[i].Trim()) == item_id)
{
item_id_index = i;
break;
}
}
for(int a = 0; a < segment.data.rows; a++)
{
if (segment.user_id[a] == active_user_id)
{
List<double> L_MSE_ij = new List<double>(); // i,i: item_id_index
List<int> Ia = new List<int>();// Item id index của active user a
double MSE_Min = -9999;
for (int j = 0; j < segment.data.cols - 1; j++) //Bỏ unknown item
{
if (segment.data[a, j] != 0)
{
Ia.Add(j);
Regression r = new Regression(item_id_index, j, segment);
double MSE_ij = r.MSE(segment);
L_MSE_ij.Add(MSE_ij);
if (MSE_Min == -9999 || MSE_ij < MSE_Min)
MSE_Min = MSE_ij;
}
}
List<double> L_MSE_Diag = new List<double>(); // {C*_i}_jj
double Inv_Sum_MSE_Diag = 0;
foreach (double MSE_ij in L_MSE_ij)
{
double MSE_Diag = MSE_ij - segment.Correlation_Avg * MSE_Min;
L_MSE_Diag.Add(MSE_Diag);
Inv_Sum_MSE_Diag += 1 / MSE_Diag;
}
int[] _Ia = Ia.ToArray();
double[] _L_MSE_Diag = L_MSE_Diag.ToArray();
double predict_ai = 0;
for(int j = 0; j < _Ia.Length; j++)
{
double wji = 1 / (_L_MSE_Diag[j] * Inv_Sum_MSE_Diag);
Regression r = new Regression(item_id_index, _Ia[j], segment);
predict_ai += wji * r.ExpertScore(segment.data[a, _Ia[j]], segment);
}
return predict_ai;
}
}
return 0;
}
public bool IsChildOf(Segment other)
{
return (budget.id == other.budget.id || other.budget.id == 0)
&& (companion.id == other.companion.id || other.companion.id == 0)
//&& (familiarity.id == other.familiarity.id || other.familiarity.id == 0)
//&& (mood.id == other.mood.id || other.mood.id == 0)
//&& (temperature.id == other.temperature.id || other.temperature.id == 0)
//&& (travelLength.id == other.travelLength.id || other.travelLength.id == 0)
&& (weather.id == other.weather.id || other.weather.id == 0);
}
private void btnTestML_Click(object sender, EventArgs e)
{
Stopwatch swglobal = new Stopwatch();
Stopwatch swtemp = new Stopwatch();
swglobal.Start();
swtemp.Start();
txtLogResample.AppendText("Process started: " + DateTime.Now);
txtLogResample.AppendText("\r\nReading Data from Cube ... ");
//Get data
Segment segment = new Segment();
string mdx = "with "
+ "member Measures.[Ratings AVG] as"
+ "( "
+ "[Measures].[Rating]/[Measures].[Ratings100k Count]"
+ ")"
+ "select "
+ "[User100k].[Id].Members on rows, "
+ "[Movies100k].[Id].Members on columns "
+ "from Movielens "
+ "where Measures.[Ratings AVG]";
DataTable result = DbHelper.RunMDXWithDataTable(mdx);
segment.data = new Matrix(result.Rows.Count - 1, result.Columns.Count - 2);
segment.user_id = new int[result.Rows.Count - 1];
segment.item_id = new string[result.Columns.Count - 2];
segment.avgRatingByItem = new double[result.Columns.Count - 2];
segment.avgRatingByUser = new double[result.Rows.Count - 1];
//item
for (int i = 0; i < segment.item_id.Length; i++)
{
segment.item_id[i] = result.Columns[i + 2].ColumnName.Trim();
segment.avgRatingByItem[i] = Convert.ToDouble(result.Rows[0][i + 2] == "" ? 0 : result.Rows[0][i + 2]);
}
for (int i = 0; i < segment.data.rows - 1; i++)
{
segment.user_id[i] = Convert.ToInt32(result.Rows[i + 1][0]);
segment.avgRatingByUser[i] = Convert.ToDouble(result.Rows[i + 1][1] == "" ? 0 : result.Rows[i + 1][1]);
for (int j = 0; j < segment.data.cols; j++)
segment.data[i, j] = Convert.ToDouble(result.Rows[i + 1][j + 2] == "" ? 0 : result.Rows[i + 1][j + 2]);
}
swtemp.Stop();
txtLogResample.AppendText(swtemp.Elapsed.ToString());
txtLogResample.AppendText("\r\nStart resampling (this may take a long time) ... ");
swtemp.Start();
List<Sample> resamples = Sample.Resampling(segment);
swtemp.Stop();
txtLogResample.AppendText(swtemp.Elapsed.ToString());
txtLogResample.AppendText("\r\nExporting result ...");
swtemp.Start();
int count = 0;
foreach (Sample sample in resamples)
{
count++;
sample.Export("D:\\Temp\\TestResampling\\" + count + ".xlsx");
}
swtemp.Stop();
txtLogResample.AppendText(swtemp.Elapsed.ToString());
swglobal.Stop();
txtLogResample.AppendText("\r\nProcess ended: " + DateTime.Now + " - Total time: " + swglobal.Elapsed.ToString());
/*
//Evaluate
double performamce_segment = 0;
double correlation_avg_segment = 0;
foreach (Sample sample in resamples)
{
double temp_corr_avg_segment = sample.Train();
correlation_avg_segment += temp_corr_avg_segment;
performamce_segment += sample.Test(temp_corr_avg_segment);
}
MessageBox.Show("Done");
txtPerformance.Text = Convert.ToString(performamce_segment / 10);
txtCorrelation.Text = Convert.ToString(correlation_avg_segment / 10);
*/
}
/*
* IEquatable
*/
public bool Equals(Segment other)
{
//Check whether the compared object is null.
if (Object.ReferenceEquals(other, null)) return false;
//Check whether the compared object references the same data.
if (Object.ReferenceEquals(this, other)) return true;
//Check whether the products' properties are equal.
return budget.id.Equals(other.budget.id)
&& companion.id.Equals(other.companion.id)
//&& familiarity.id.Equals(other.familiarity.id)
//&& mood.id.Equals(other.mood.id)
//&& temperature.id.Equals(other.temperature.id)
//&& travelLength.id.Equals(other.travelLength.id)
&& weather.id.Equals(other.weather.id);
}
public static Segment[] GetCandidates()
{
DataTable data = DbHelper.RunScriptsWithTable(string.Format("select * from segments order by performance asc", "Data Warehouse"), "Data Warehouse");
Segment[] candidates = new Segment[data.Rows.Count];
for (int i = 0; i < data.Rows.Count; i++)
{
Segment obj = new Segment();
obj.id = Convert.ToInt32(data.Rows[i][0]);
//Time
string period_of_day = Convert.ToString(data.Rows[i][1]);
string period_of_week = Convert.ToString(data.Rows[i][2]);
string season = Convert.ToString(data.Rows[i][3]);
obj.time = new Time(period_of_day, period_of_week, season);
obj.budget = new Budget(Convert.ToInt32(data.Rows[i][4]));
obj.companion = new Companion(Convert.ToInt32(data.Rows[i][5]));
obj.weather = new Weather(Convert.ToInt32(data.Rows[i][6]));
obj.Performance = Convert.ToDouble(data.Rows[i][7]);
obj.Correlation_Avg = Convert.ToDouble(data.Rows[i][8]);
candidates[i] = obj;
}
return candidates;
}
public static List<Segment> GetAllSegment()
{
List<Segment> result = new List<Segment>();
foreach(Time time in Time.GetAll())
foreach(Budget budget in Budget.GetAllData())
foreach(Companion companion in Companion.GetAllData())
//foreach(Familiarity familiarity in Familiarity.GetAllData())
//foreach(Mood mood in Mood.GetAllData())
//foreach(Temperature temperature in Temperature.GetAllData())
//foreach(TravelLength travelLength in TravelLength.GetAllData())
foreach(Weather weather in Weather.GetAllData())
{
if (time.period_of_day == Time.Period_Of_Day.All && time.period_of_week == Time.Period_Of_Week.All && time.season == Time.Season.All && budget.id == 0 && companion.id == 0 && weather.id == 0)
continue;
else
{
Segment segment = new Segment();
segment.time = time;
segment.budget = budget;
segment.companion = companion;
//segment.familiarity = familiarity;
//segment.mood = mood;
//segment.temperature = temperature;
//segment.travelLength = travelLength;
segment.weather = weather;
segment.GetData();
if (segment.data.CountCells() > 100)
result.Add(segment);
}
}
return result;
}
/*
* Static Methods
*/
public static Segment GetRoot()
{
Segment root = new Segment();
string mdx = "with member Measures.[Avg_Ratings] as "
+ "([Measures].[Sum_Ratings]/[Measures].[Count_Ratings]) "
+ "select "
+ "[Dim Place].[Place Key].Members on columns, "
+ "[Dim User].[User Key].Members on rows "
+ "from [Travel H2V DW]"
+ "where Measures.[Avg_Ratings]";
DataTable result = DbHelper.RunMDXWithDataTable(mdx);
root.data = new Matrix(result.Rows.Count - 1, result.Columns.Count - 2);
root.user_id = new int[result.Rows.Count - 1];
root.item_id = new string[result.Columns.Count - 2];
root.avgRatingByItem = new double[result.Columns.Count - 2];
root.avgRatingByUser = new double[result.Rows.Count - 1];
//item
for (int i = 0; i < root.item_id.Length; i++)
{
root.item_id[i] = result.Columns[i + 2].ColumnName.Trim();
root.avgRatingByItem[i] = Convert.ToDouble(result.Rows[0][i + 2] == "" ? 0 : result.Rows[0][i + 2]);
}
for (int i = 0; i < root.data.rows; i++)
{
root.user_id[i] = Convert.ToInt32(result.Rows[i + 1][0]);
root.avgRatingByUser[i] = Convert.ToDouble(result.Rows[i + 1][1] == "" ? 0 : result.Rows[i + 1][1]);
for (int j = 0; j < root.data.cols; j++)
root.data[i, j] = Convert.ToDouble(result.Rows[i + 1][j + 2] == "" ? 0 : result.Rows[i + 1][j + 2]);
}
return root;
}
public static double Correlation_Avg(Segment segment) // Training Phase
{
double sum = 0;
int count = 0;
for (int i = 0; i < segment.item_id.Length - 3; i++)
for (int j = i + 1; j < segment.item_id.Length - 2; j++)
for (int k = j + 1; k < segment.item_id.Length - 1; k++)
{
double temp = Correlation(i, j, k, segment);
if (Double.IsInfinity(temp))
temp = Correlation(i, j, k, segment);
if (!Double.IsNaN(temp) && !Double.IsInfinity(temp))
{
sum += temp;
count++;
}
}
/*
for (int s = 0; s < segment.item_id.Length; s++)
{
Random rand = new Random();
int i = 1, j = 2, k = 3;
while (i != j && i != k && j != k)
{
i = rand.Next(0, segment.item_id.Length - 1);
j = rand.Next(0, segment.item_id.Length - 1);
k = rand.Next(0, segment.item_id.Length - 1);
}
double temp = Correlation(i, j, k, segment);
if (Double.IsInfinity(temp))
temp = Correlation(i, j, k, segment);
if (!Double.IsNaN(temp) && !Double.IsInfinity(temp))
{
sum += temp;
count++;
}
}*/
return sum / count;
}
public static bool GetStrongSegments()
{
//List<Segment> AllSegment = Segment.GetAllSegment();
Segment root = Segment.GetRoot();
double sum_Correlation_Avg_root = 0;
int count_Correlation_Avg_root = 0;
DbHelper.RunScripts("truncate table segments", "Data Warehouse");
//foreach (Segment segment in AllSegment)
//{
foreach(Time time in Time.GetAll())
foreach(Budget budget in Budget.GetAllData())
foreach(Companion companion in Companion.GetAllData())
foreach (Weather weather in Weather.GetAllData())
{
if (time.period_of_day == Time.Period_Of_Day.All && time.period_of_week == Time.Period_Of_Week.All && time.season == Time.Season.All && budget.id == 0 && companion.id == 0 && weather.id == 0)
continue;
else
{
Segment segment = new Segment();
segment.time = time;
segment.budget = budget;
segment.companion = companion;
//segment.familiarity = familiarity;
//segment.mood = mood;
//segment.temperature = temperature;
//segment.travelLength = travelLength;
segment.weather = weather;
segment.GetData();
if (segment.data.CountCells() < 10)
continue;
//result.Add(segment);
//Resample segment
double performamce_segment = 0;
double performance_root = 0;
double correlation_avg_segment = 0;
List<Sample> resamples = Sample.Resampling(segment);
foreach (Sample sample in resamples)
{
double temp_corr_avg_segment = sample.Train();
if (!Double.IsNaN(temp_corr_avg_segment))
{
correlation_avg_segment += temp_corr_avg_segment;
performamce_segment += sample.Test(temp_corr_avg_segment);
}
Sample sample_root = new Sample(sample.Evaluation, sample.Training_root);
double temp_corr_root = sample_root.Train();
if (!Double.IsNaN(temp_corr_root))
{
sum_Correlation_Avg_root += temp_corr_root;
count_Correlation_Avg_root += 1;
performance_root += sample_root.Test(sum_Correlation_Avg_root / count_Correlation_Avg_root);
}
}
//if (performamce_segment < performance_root)
//if (!Double.IsNaN(performamce_segment) && !Double.IsNaN(correlation_avg_segment))
if (Double.IsNaN(performamce_segment))
performamce_segment = -9999;
if (Double.IsNaN(correlation_avg_segment))
correlation_avg_segment = -9999;
if (Double.IsNaN(performance_root))
performance_root = -9999;
try
{
DbHelper.RunScripts(string.Format("pr_insertSegment "
+ "'" + segment.time.period_of_day.ToString() + "'"
+ ", '" + segment.time.period_of_week.ToString() + "'"
+ ", '" + segment.time.season.ToString() + "'"
+ ", " + segment.budget.id
+ ", " + segment.companion.id
+ ", " + segment.weather.id
+ ", " + performamce_segment / 10
+ ", " + correlation_avg_segment / 10
+ ", " + performance_root / 10
+ ", " + sum_Correlation_Avg_root / count_Correlation_Avg_root)
, "Data Warehouse");
}
catch (Exception ex)
{ }
}
//Remove segment "child" and have performance less than its parents
/*
Segment[] candidates = Segment.GetCandidates();
for (int i = 0; i < candidates.Length - 1; i++)
{
for (int j = i + 1; j < candidates.Length; j++)
{
if (candidates[j].IsChildOf(candidates[i]))
{
DbHelper.RunScripts(string.Format("delete from segments where id = " + candidates[j].id), "Data Warehouse");
}
}
}*/
}
DbHelper.RunScripts(string.Format("pr_insertSegment "
+ "'All', 'All', 'All'"
+ ", " + 0
+ ", " + 0 + ", " + 0
+ ", " + 9999
+ ", " + sum_Correlation_Avg_root / count_Correlation_Avg_root), "Data Warehouse");
return true;
}
