public HlaMsr1 GetGroundInstance(string name, ref HashSet <string> warningSet)
{
HlaMsr1 hla = (HlaMsr1)GetGroundOrAbstractInstance(name, ref warningSet);
SpecialFunctions.CheckCondition(hla.IsGround, "Hla is not ground. " + name);
return(hla);
}
//!!!Perforance ideas: Could the backoff models be applied better than just one at a time until one works?
//!!!Perforance ideas: Does the web use load all the models on every use? Is this a problem?
private static void RemoveRedundantTerms(ref List <KeyValuePair <string, string> > termAndInParenList, ref HashSet <string> warningSet)
{
termAndInParenList.Sort((termAndInParen1, termAndInParen2) => termAndInParen1.Key.Length.CompareTo(termAndInParen2.Key.Length));
for (int i = termAndInParenList.Count - 1; i > 0; --i)
{
string keyI = termAndInParenList[i].Key;
for (int j = 0; j < i; ++j)
{
string keyJ = termAndInParenList[j].Key;
if (HlaMsr1.IsMoreGeneralOrEqualThan(keyJ, keyI))
{
if (keyJ == keyI)
{
termAndInParenList[j] = new KeyValuePair <string, string>(keyJ, termAndInParenList[j].Value + "/" + termAndInParenList[1].Value);
termAndInParenList.RemoveAt(i);
warningSet.Add("Two terms combined because they are equal at the generality used by the model.");
break; // really break, not continue;
}
throw new GeneralizingTermException(keyJ, keyI, string.Format("Term {0} generalizes term {1}", keyJ, keyI));
}
}
}
}
public static IEnumerable <PidAndHlaSet> GetEnumerationSparse(TextReader inputTextReader)
{
string previousPid = null;
Dictionary <string, List <HlaMsr1> > classToHlaList = null;
HashSet <string> warningSet = null;
foreach (var pidAndHlaRecord in SpecialFunctions.ReadDelimitedFile(inputTextReader, new { pid = "", hla = "" }, new char[] { '\t' }, true))
{
string pid = pidAndHlaRecord.pid;
if (previousPid != pid)
{
if (previousPid != null)
{
PidAndHlaSet pidAndHlaSet = CreatePidAndHlaSet(previousPid, classToHlaList, warningSet);
yield return(pidAndHlaSet);
}
previousPid = pid;
classToHlaList = new Dictionary <string, List <HlaMsr1> >();
warningSet = new HashSet <string>();
}
HlaMsr1 hlaMsr1 = (HlaMsr1)HlaMsr1Factory444.GetGroundOrAbstractInstance(pidAndHlaRecord.hla, ref warningSet);
List <HlaMsr1> hlaList = classToHlaList.GetValueOrDefault(hlaMsr1.ClassName);
hlaList.Add(hlaMsr1);
}
if (previousPid != null)
{
PidAndHlaSet pidAndHlaSet = CreatePidAndHlaSet(previousPid, classToHlaList, warningSet);
yield return(pidAndHlaSet);
}
}
private void SetGroundOnHla(List <KeyValuePair <string, string> > termList, HlaMsr1 hlaMsrMsr1)
{
if (termList.Count > 1)
{
hlaMsrMsr1._isGround = false;
}
else
{
SetGroundOnSingleTermHla(hlaMsrMsr1);
}
}
private void SetGroundOnSingleTermHla(HlaMsr1 hlaMsrMsr1)
{
int hlaLength = ClassNameToHlaLength[hlaMsrMsr1.ClassName];
switch (hlaLength)
{
case 4:
{
if (hlaMsrMsr1.Name.Length == 1)
{
hlaMsrMsr1._isGround = false;
}
else if (HlaMsr1Factory.TrimToTwoList.Contains(hlaMsrMsr1.Name))
{
hlaMsrMsr1._isGround = true;
}
else
{
hlaMsrMsr1._isGround = hlaMsrMsr1.Name.Length == 5;
}
}
break;
case 2:
{
if (hlaMsrMsr1.Name.Length == 1)
{
hlaMsrMsr1._isGround = false;
}
else
{
hlaMsrMsr1._isGround = hlaMsrMsr1.Name.Length == 3;
}
}
break;
case 0:
{
hlaMsrMsr1._isGround = hlaMsrMsr1.Name.Length == 1;
}
break;
default:
SpecialFunctions.CheckCondition(false, "Don't have code for hlaLength " + hlaLength.ToString());
break;
}
}
public Hla GetGroundOrAbstractInstance(string name, ref HashSet <string> warningSet)
{
string soFar = name.ToUpperInvariant();
var keyAndInParenListX = CreateCanonicalTermList(soFar, name, ref warningSet);
var keyAndOrigList =
(from kv in keyAndInParenListX
let k = kv.Key
let inParen = kv.Value
let ori = (inParen == "") ? k : string.Format("{0}({1})", k, inParen)
select new KeyValuePair <string, string>(k, ori)).ToList();
RemoveRedundantTerms(ref keyAndOrigList, ref warningSet);
keyAndOrigList.Sort(SortTermsAlpha);
HlaMsr1 hlaMsrMsr1 = new HlaMsr1(keyAndOrigList);
SetGroundOnHla(keyAndOrigList, hlaMsrMsr1);
return(hlaMsrMsr1);
}
private IEnumerable <HlaMsr1> GeneralizationList(HlaMsr1 groundHla)
{
return(groundHla.GeneralizationList(HlaMsr1Factory));
}
private IEnumerable <KeyValuePair <HlaMsr1, double> > ExpandHla(TableInfo tableInfo, HlaMsr1 hlaAbstractOrGround, LinkedList1 <HlaMsr1> linkedList1)
{
HashSet <HlaMsr1> groundSet = new HashSet <HlaMsr1>();
foreach (HlaMsr1 term in hlaAbstractOrGround.TermList(tableInfo.HlaMsr1Factory))
{
List <HlaMsr1> groundHlaList;
if (tableInfo.AbstractHlaToGroundHlaList.TryGetValue(term, out groundHlaList))
{
if (groundHlaList.Count == 1 && groundHlaList.First().Equals(term))
{
groundSet.AddNew(term);
}
else
{
foreach (HlaMsr1 ground in groundHlaList)
{
groundSet.AddNew(ground);
}
}
}
}
if (groundSet.Count == 0)
{
yield break;
}
List <Dictionary <HlaMsr1, double> > rowsOfInterest = tableInfo.PullOutTheRowsOfInterest(linkedList1);
//!!!for each list of rowsOfInterest we could cache the sum of exp's to speed things up
//!!!This could be made faster by giving a serial number to each HLA and then doing the calcuations in arrays in which the serial number is the index.
Dictionary <HlaMsr1, double> hlaToTotal = new Dictionary <HlaMsr1, double>();
foreach (Dictionary <HlaMsr1, double> hlaToWeight in rowsOfInterest)
{
foreach (KeyValuePair <HlaMsr1, double> hlaAndWeight in hlaToWeight)
{
hlaToTotal[hlaAndWeight.Key] = hlaToTotal.GetValueOrDefault(hlaAndWeight.Key) + hlaAndWeight.Value;
}
}
Dictionary <HlaMsr1, double> hlaToExpTotal = new Dictionary <HlaMsr1, double>();
double totalOfExpsPlus1 = 1;
foreach (KeyValuePair <HlaMsr1, double> hlaAndTotal in hlaToTotal)
{
double exp = Math.Exp(hlaAndTotal.Value);
totalOfExpsPlus1 += Math.Exp(hlaAndTotal.Value);
hlaToExpTotal.Add(hlaAndTotal.Key, exp);
}
foreach (HlaMsr1 hlaGround in groundSet)
{
double prob = hlaToExpTotal[hlaGround] / totalOfExpsPlus1;
yield return(new KeyValuePair <HlaMsr1, double>(hlaGround, prob));
}
}
private IEnumerable <KeyValuePair <HlaMsr1, KeyValuePair <double, bool> > > ExpandHla(HlaMsr1 hlaAbstractOrGround, LinkedList1 <HlaMsr1> linkedList1)
{
//Reference: http://en.wikipedia.org/wiki/Multinomial_logit#Model
//Dictionary<string, Dictionary<HlaMsr1, double>> predictorNameToTargetHlaToWeight = Ethnicity.ClassNamePredictorNameOrInterceptToTargetToWeight[hlaAbstractOrGround.ClassName];
EClass eclass = Ethnicity.HlaClassNameToEClass[hlaAbstractOrGround.ClassName];
bool usedLowerResModel = false;
foreach (LinkedList1 <int> hlaLengthList in eclass.HlaLengthListSorted())
{
bool anyReturned = false;
TableInfo tableInfo = eclass.HlaLengthListToTableInfo[hlaLengthList];
foreach (var hlaAndProb in ExpandHla(tableInfo, hlaAbstractOrGround, linkedList1))
{
anyReturned = true;
yield return(new KeyValuePair <HlaMsr1, KeyValuePair <double, bool> >(hlaAndProb.Key, new KeyValuePair <double, bool>(hlaAndProb.Value, usedLowerResModel)));
}
if (anyReturned)
{
yield break; // really yield break, not yield return or return;
}
usedLowerResModel = true;
}
throw new HlaNotInModelException(hlaAbstractOrGround.Name, string.Format("Can't find {0} or any prefix in any model", hlaAbstractOrGround.ToString(/*withParen*/ true)));
}
