/// <summary>
/// Assumption: input data already sorted by most recent first.
/// </summary>
public static double[] GetTransactions_Max_PRIOR_CREDIT_LIMIT_AMT(int len, DateTime predictionStartDate, List <DataTransactions> list, int days)
{
double[] nbs = NnRow.CreateArray(len);
if (uNet.IsNullOrEmpty(list))
{
return(nbs);
}
var tempList = new List <DataTransactions>(list);
DateTime Start = predictionStartDate;
DateTime End = predictionStartDate;
for (int i = 0; i < nbs.Length && 0 < tempList.Count; i++)
{
End = Start;
Start = End - TimeSpan.FromDays(days);
var shortList = tempList.FindAll(e => Start < e.TRANSACTION_DTTM && e.TRANSACTION_DTTM <= End).ToList();
if (shortList.Count <= 0)
{
continue;
}
double max = shortList.Max(e => e.PRIOR_CREDIT_LIMIT_AMT);
nbs[i] = max;
}
nbs = NormalizeToOne(nbs, 0, 39709); // -48~+18044 for A, 0~+5427 for B, -5709~+39709 for C,
return(nbs);
}
/// <summary>
/// Assumption: input data already sorted by most recent first.
/// </summary>
public static double[] GetFacturationCashBalance(int len, DateTime predictionStartDate, List <DataFacturation> list, int days,
double min, double max)
{
double[] nbs = NnRow.CreateArray(len);
if (uNet.IsNullOrEmpty(list))
{
return(nbs);
}
var tempList = new List <DataFacturation>(list);
DateTime Start = predictionStartDate;
DateTime End = predictionStartDate;
for (int i = 0; i < nbs.Length && 0 < tempList.Count; i++)
{
End = Start;
Start = End - TimeSpan.FromDays(days);
var shortList = tempList.FindAll(e => Start < e.PERIODID_MY && e.PERIODID_MY <= End).ToList();
if (shortList.Count <= 0)
{
continue;
}
double val = shortList.Max(e => e.CashBalance);
nbs[i] = val;
}
nbs = NormalizeToOne(nbs, min, max);
return(nbs);
}
private static void Init_Categories(List <DataTransactions> list, DateTime predictionStartDate, int nbDaysMostRecent,
List <string> categories, out double[] nbs, out List <DataTransactions> shortList)
{
nbs = NnRow.CreateArray(categories.Count);
DateTime Start = predictionStartDate;
DateTime End = predictionStartDate;
Start = End - TimeSpan.FromDays(nbDaysMostRecent);
shortList = list.FindAll(e => Start < e.TRANSACTION_DTTM && e.TRANSACTION_DTTM <= End).ToList();
}
public static double[] GetFacturationCurrentTotalBalance(List <DataFacturation> list)
{
double[] nbs = NnRow.CreateArray(5); if (uNet.IsNullOrEmpty(list))
{
return(nbs);
}
var vals = new List <double>();
for (int i = 0; i < list.Count; i++)
{
vals.Add(list[i].CurrentTotalBalance);
}
CalculateStatsAndCopyToArray(nbs, vals);
nbs = NormalizeToOne(nbs);
return(nbs);
}
public static double[] GetPaiementsStats(List <DataPaiements> list)
{
double[] nbs = NnRow.CreateArray(5); if (uNet.IsNullOrEmpty(list))
{
return(nbs);
}
var vals = new List <double>();
for (int i = 0; i < list.Count; i++)
{
vals.Add(list[i].TRANSACTION_AMT);
}
CalculateStatsAndCopyToArray(nbs, vals);
nbs = NormalizeToOne(nbs);
return(nbs);
}
public static List <NnRow> outputs = new List <NnRow>(); // contains all rows of numbers to write for N.Network
/// <summary>
/// Convert a curation csv file, containing such as textual biz names and addresses, to a csv file with only matched results in double.
/// Can return null if the matched file already exist, or if something went wrong during calculations.
/// </summary>
/// <param name="CsvPath"></param>
/// <returns></returns>
public static List <NnRow> AnalyzeAndCreateColumnsForNNetwork(bool trainNotTest, bool useFull, bool loadBin)
{
DicFact dicFact; DicPaie dicPaie; DicPerf dicPerf; DicTran dicTran;
if (loadBin)
{
Utils.GetDataFromBinFiles(out dicFact, out dicPaie, out dicPerf, out dicTran);
}
else
{
Utils.GetDataFromCsvFiles(out dicFact, out dicPaie, out dicPerf, out dicTran, trainNotTest, useFull);
}
//ExportDefaultedTransactions(dicPerf, dicTran); // debugOnly
// create csv files containing that embed text in numerical values, by matching parts of business name and address
int nbRows = dicPerf.Keys.Count;
var clients = dicPerf.Keys.ToList();
var rowArray = new NnRow[nbRows]; // we use an array here, because a list isn't compatible with parallelism
/**/
// Using single CPU core, good for debugging
for (int i = 0; i < nbRows; i++)
{
rowArray[i] = CalculateRow(clients[i], dicFact, dicPaie, dicPerf, dicTran);
}
/*
* // Using multiple CPU cores, use only if there's a lot to process
* Parallel.For(0, nbRows, new ParallelOptions { MaxDegreeOfParallelism = 4 }, i =>
* {
* rowArray[i] = CalculateRow(clients[i], dicFact, dicPaie, dicPerf, dicTran);
* });
* /**/
List <NnRow> dataset = rowArray.ToList();
dataset.RemoveAll(e => null == e);
return(dataset);
}
/// <summary>
/// Assumption: input data already sorted by most recent first.
/// </summary>
public static double[] GetTransactions(int len, DateTime predictionStartDate, List <DataTransactions> list, int days)
{
double[] nbs = NnRow.CreateArray(len);
if (uNet.IsNullOrEmpty(list))
{
return(nbs);
}
var tempList = new List <DataTransactions>(list);
DateTime Start = predictionStartDate;
DateTime End = predictionStartDate;
for (int i = 0; i < nbs.Length && 0 < tempList.Count; i++)
{
End = Start;
Start = End - TimeSpan.FromDays(days);
var shortList = tempList.FindAll(e => Start < e.TRANSACTION_DTTM && e.TRANSACTION_DTTM <= End).ToList();
double sum = shortList.Sum(e => e.TRANSACTION_AMT);
nbs[i] = sum;
}
nbs = NormalizeToOne(nbs);
return(nbs);
}
/// <summary>
/// Assumption: input data already sorted by most recent first.
/// </summary>
public static double[] GetSpendsAndPaymentsAfterStatement(int len, DateTime predictionStartDate, List <DataFacturation> rowsFact,
List <DataPaiements> rowsPaie, List <DataTransactions> rowsTran, int days)
{
var allStats = new List <double>();
DateTime Start = predictionStartDate;
DateTime End = predictionStartDate;
for (int i = 0; i < len; i++)
{
End = Start;
Start = End - TimeSpan.FromDays(days);
var shortTran = rowsTran.FindAll(e => Start < e.TRANSACTION_DTTM && e.TRANSACTION_DTTM <= End).ToList();
if (shortTran.Count <= 0)
{
continue;
}
shortTran = shortTran.OrderByDescending(e => e.TRANSACTION_DTTM).ToList();
var shortPaie = null == rowsPaie ? new List <DataPaiements>() :
rowsPaie.FindAll(e => Start < e.TRANSACTION_DTTM && e.TRANSACTION_DTTM <= End).ToList();
if (shortPaie.Count <= 0)
{
continue;
}
shortPaie = shortPaie.OrderByDescending(e => e.TRANSACTION_DTTM).ToList();
var shortFact = null == rowsFact ? new List <DataFacturation>() :
rowsFact.FindAll(e => (Start - TimeSpan.FromDays(days + 90)) < e.PERIODID_MY && e.PERIODID_MY <= Start).ToList();
shortFact = shortFact.OrderByDescending(e => e.PERIODID_MY).ToList();
double lastBalance = shortFact.Count <= 0? 0: shortFact[0].CurrentTotalBalance;
List <(DateTime date, double val)> chronological = new List <(DateTime date, double val)>();
//chronological.Add((new DateTime(1000, 1, 1), lastBalance));
foreach (var e in shortTran)
{
chronological.Add((e.TRANSACTION_DTTM, e.TRANSACTION_AMT));
}
foreach (var e in shortPaie)
{
chronological.Add((e.TRANSACTION_DTTM, -e.TRANSACTION_AMT));
}
chronological = chronological.OrderBy(e => e.date).ToList(); // oldest first
var vals = new List <double>();
foreach (var e in chronological)
{
lastBalance += e.val;
vals.Add(lastBalance);
}
double[] stats = NnRow.CreateArray(4);
CalculateStatsAndCopyToArray(stats, vals);
stats = NormalizeToOne(stats, -60000, 60000);
allStats.AddRange(stats);
}
// !!! here we want a neutral balance to be 0.5, so we can have both +- values in the 0.0 ~ 1.0 output range
double neutral = 0.5;
double[] nbs = NnRow.CreateArray(len * 4, neutral);
for (int i = 0; i < nbs.Length && i < allStats.Count; i++)
{
nbs[i] = allStats[i];
}
return(nbs);
}