/* ======================================================================== *\ ! ! * ! * This file is part of MARS, the MAGIC Analysis and Reconstruction ! * Software. It is distributed to you in the hope that it can be a useful ! * and timesaving tool in analysing Data of imaging Cerenkov telescopes. ! * It is distributed WITHOUT ANY WARRANTY. ! * ! * Permission to use, copy, modify and distribute this software and its ! * documentation for any purpose is hereby granted without fee, ! * provided that the above copyright notice appear in all copies and ! * that both that copyright notice and this permission notice appear ! * in supporting documentation. It is provided "as is" without express ! * or implied warranty. ! * ! ! ! Author(s): Thomas Bretz, 04/2004 ! ! Copyright: MAGIC Software Development, 2000-2008 ! ! \* ======================================================================== */ ///////////////////////////////////////////////////////////////////////////// // // MDataPhrase // // A MDataPhrase is a wrapper for TFormula. It supports access to data- // members and/or member functions acessible from the parameter list // via MDataMember. It supports access to elements of a MMatrix through // the parameter list via MDataElement and it sipports variables set // by SetVariables via MDataValue. // // The parsing is done by TFormula. For more information which functions // are supported see TFormula, TFormulaPrimitives and TFormulaMathInterface. // // The support is done by replacing the parameters with access to the // parameter list by parameters like [0], [1],... When evaluating // the TFormula first the parameters are evaluated and passed to // TFormula. Each parameter is evaluated only once and, if necessary, // passed more than once to TFormula. To store the MData* classes used // for parameter access a TObjArray is used. Its advantage is, that // it has an UncheckedAt function which saves some time, because // no obsolete sanity checks must be done accessing the array. // // Because everything supported by TFormula is also supported by // MDataPhrase also conditional expression are supported. // // For supported functions see TFormulaPrimitive and TMathInterface. // // Examples: // // Gaus, Gausn, Landau, Landaun, Pol0-Pol10, Pow2-Pow5 // // In the constructor you can give rule, like // "HillasSource.fDist / MHillas.fLength" // Where MHillas/HillasSource is the name of the parameter container in // the parameter list and fDist/fLength is the name of the data members // in the containers. The result will be fDist divided by fLength. // // In case you want to access a data-member which is a data member object // you can acces it with (Remark: it must derive from MParContainer): // "MCameraLV.fPowerSupplyA.fVoltagePos5V" // (THIS FEATURE IS CURRENTLY NOT SUPPORTED) // // You can also use parantheses: // "HillasDource.fDist / (MHillas.fLength + MHillas.fWidth)" // // Additional implementations: // // isnan(x) return 1 if x is NaN (Not a Number) otherwise 0 // finite(x) return 1 if the number is a valid double (not NaN, inf) // // NaN (Not a Number) means normally a number which is to small to be // stored in a floating point variable (eg. 0 #include #include #include #if ROOT_VERSION_CODE >= ROOT_VERSION(5,26,00) #if ROOT_VERSION_CODE < ROOT_VERSION(6,00,00) #include #else #include #endif #endif #include "MLog.h" #include "MLogManip.h" #include "MArrayD.h" #include "MDataValue.h" #include "MDataMember.h" #include "MDataElement.h" ClassImp(MDataPhrase); using namespace std; // -------------------------------------------------------------------------- // // Check for the existance of the expression [idx] in the string // phrase. If existing a corresponding new MDataValue is added to // fMembers and index is increased by one. // // This makes the use of user-defined variables in the phrase possible. // Int_t MDataPhrase::CheckForVariable(const TString &phrase, Int_t idx) { TString mods; TArrayI pos; while (1) { // \\A: matches the beginning of the string like ^ does // \\Z: matches the end of the string like $ does // \\W: matches any non-word character [^a-zA-Z_0-9] TPRegexp reg(Form("(\\A|\\W)\\[0*%d\\](\\W|\\Z)", idx)); if (reg.Match(phrase, mods, 0, 130, &pos)==0) break; // [idx] already existing. Add a corresponding MDataValue fMembers.AddLast(new MDataValue(0, idx)); idx++; } return idx; } // -------------------------------------------------------------------------- // // Replace all expressions expr (found by a regular expression \\b%s\\b // with %s being the expression) in the string phrase by [idx]. // // The length of [idx] is returned. // Int_t MDataPhrase::Substitute(TString &phrase, const TString &expr, Int_t idx) const { const TString arg = Form("[%d]", idx); TPRegexp reg(expr); TString mods; TArrayI pos; Int_t p = 0; while (1) { // check whether argument is found if (reg.Match(phrase, mods, p, 130, &pos)==0) break; // Replace expression by argument [idx] phrase.Replace(pos[0], pos[1]-pos[0], arg, arg.Length()); // Jump behind the new string which was just inserted p = pos[0]+arg.Length(); } return arg.Length(); } TString MDataPhrase::Parse(TString phrase) { // This is for backward compatibility with MDataChain phrase.ReplaceAll("gRandom->", "TRandom::"); phrase.ReplaceAll("kRad2Deg", "TMath::RadToDeg()"); phrase.ReplaceAll("kDeg2Rad", "TMath::DegToRad()"); phrase.ReplaceAll(" ", ""); int idx=0; int p =0; TString mods; TArrayI pos; /* // Find all functions... // The first \\w+ should also allow :: TPRegexp reg = TPRegexp("[[:word:]:]+\\([^()]*\\)"); while (1) { //idx = CheckForVariables(phrase, idx); if (reg.Match(phrase, mods, p, 130, &pos)==0) break; const Int_t len = pos[1]-pos[0]; // This is the full function with its argument(s) const TString term = phrase(pos[0], len); // Now get rid of the argument(s) const TPRegexp reg3("\\([^()]+\\)"); TString func(term); reg3.Substitute(func, ""); // Seems to be a special case whic is not handles by // TFormulaPrimitive by known by TFormula if (func.BeginsWith("TRandom::")) { p = pos[0]+pos[1]; continue; } // check whether the function is available if (TFormulaPrimitive::FindFormula(func)) { p = pos[0]+pos[1]; continue; } cout << "Unknown: " << func << endl; // p = pos[0]+pos[1]; return; } */ p = 0; // Find all data-members in expression such as // MTest.fDataMember. Because also floating point // numbers can contain a dot the result has to // be checked carefully // \\w: matches any word character [a-zA-Z_0-9] TPRegexp reg = TPRegexp("\\w+[.]\\w+"); TPRegexp ishex("^0x[[:xdigit:]]+$"); while (1) { // If some indices are already existing // initialize them by a flexible MDataValue idx = CheckForVariable(phrase, idx); // Check whether expression is found if (reg.Match(phrase, mods, p, 130, &pos)==0) break; // Get expression from phrase const TString expr = phrase(pos[0], pos[1]-pos[0]); // Also hex-numbers and floats fullfill our condition... // FIXME: WHY HEX NUMBERS? if (!expr(ishex).IsNull() || expr.IsFloat()) { p = pos[1]; continue; } // Add a corresponding MDataMember to our list fMembers.AddLast(new MDataMember(expr)); // Find other occurances of arg by this regexp // and start next search behind first match const TString regexp = Form("\\b%s\\b", expr.Data()); p = pos[0] + Substitute(phrase, regexp, idx); // Step forward to the next argument idx++; } p = 0; // Now check for matrix elemets as M[5] reg = TPRegexp("\\w+\\[\\d+\\]"); while (1) { // If some indices are already existing // initialize them by a MDataValue idx = CheckForVariable(phrase, idx); // Check whether expression is found if (reg.Match(phrase, mods, p, 130, &pos)==0) break; // Get expression from phrase TString expr = phrase(pos[0], pos[1]-pos[0]); // Add a corresponding MDataMember to our list fMembers.AddLast(new MDataElement(expr)); // Make the expression "Regular expression proofed" expr.ReplaceAll("[", "\\["); expr.ReplaceAll("]", "\\]"); // Find other occurances of arg by this regexp // and start next search behind first match const TString regexp = Form("\\b%s", expr.Data()); p = pos[0] + Substitute(phrase, regexp, idx); // Step forward to the next argument idx++; } /* // * HOW DO WE PROCESS THE FILTERS? // * DO WE NEED THIS FOR MData derived classes? Is there any need for it? // * MAYBE FIRST FILTERS (MF?) MUST REPLACE {name[class]} BEFORE // THE DATA PHRASSE IS CREATED? // --> MFDataPhrase must have a list of MFilter. In Process first // all self created MFilter are processed (see MF). Then // they are evaluated and the result is given to the MDataPhrase. // Can this be done using MDataMember? We replace {Name[class]} // by Name.IsExpressionTrue and we need a way that MDataPhrase // gets the coresponding pointer. // --> Alternatively we can create a MDataPhrase which allows // Pre/Processing // // We convert {Name[Class]} to Name.IsExpressionTrue. To process these // data/filters we need a request from MFDataPhrase (new virtual // memeber function?) // // {} Is alreaqdy used in ReadEnv. // // Enhance ReadEnv to allow Cut1.5 to be just a class. // // The difference between MFDataPhrase is // MFDataPhrase only knows MDataPhrase, while // MF also knows a filter-class. // p = 0; // And now we check for other phrases or filters // They are defined by a [, a pribtable character and // any numer of word characters (a-zA-Z0-9_) and a closing ] reg = TPRegexp("{[A-Za-z}\\w+(\\[[A-Za-z]\\w+\\])?}"); while (1) { // If some indices are already existing // initialize them by a MDataValue idx = CheckForVariable(phrase, idx); // Check whether expression is found if (reg.Match(phrase, mods, p, 130, &pos)==0) break; // Get expression from phrase TString expr = phrase(pos[0], pos[1]-pos[0]); // Zerlegen: {Name[Class]} // Create a new MData object of kind MData *dat = (MData*)GetNewObject(cls, MData::Class()); if (!dat) return ""; dat->SetName(name); // Add a corresponding MDataMember to our list fMembers.AddLast(dat); // Make the expression "Regular expression proofed" expr.ReplaceAll("[", "\\["); expr.ReplaceAll("]", "\\]"); // Find other occurances of arg by this regexp // and start next search behind first match p = pos[0] + Substitute(phrase, expr, idx); // Step forward to the next argument idx++; } */ // Now we have to check if there are additional indices [idx] // This is mainly important if the rule has indices only! while (1) { const Int_t newidx = CheckForVariable(phrase, idx); if (newidx == idx) break; } return phrase; } // -------------------------------------------------------------------------- // // Clear Formula and corresponding data members // void MDataPhrase::Clear(Option_t *) { fMembers.Delete(); if (!fFormula) return; delete fFormula; fFormula = NULL; } void MDataPhrase::RecursiveRemove(TObject *obj) { if (fFormula==obj) fFormula = NULL; } // -------------------------------------------------------------------------- // // Set a new phrase/rule. Returns kTRUE on sucess, kFALSE otherwise // Bool_t MDataPhrase::SetRule(const TString &rule) { Clear(); const TString txt=Parse(rule); if (txt.IsNull()) { Clear(); return kFALSE; } fFormula = new TFormula; fFormula->SetNameTitle(fName.IsNull()?"TFormula":fName.Data(), txt); fFormula->SetBit(kMustCleanup); // Must have a name otherwise all axis labels disappear like a miracle if (fFormula->Compile()) { *fLog << err << dbginf << "Syntax Error: TFormula::Compile failed..."<< endl; *fLog << " Full Rule: " << rule << endl; *fLog << " Parsed Rule: " << txt << endl; Clear(); return kFALSE; } gROOT->GetListOfFunctions()->Remove(fFormula); return kTRUE; } namespace MFastFun { // // Namespace with basic primitive functions registered by TFormulaPrimitive // all function registerd by TFormulaPrimitive can be used in TFormula // Double_t Nint(Double_t x){return TMath::Nint(x);} Double_t Sign(Double_t x){return x<0?-1:+1;} Double_t IsNaN(Double_t x){return TMath::IsNaN(x);} Double_t Finite(Double_t x){return TMath::Finite(x);} } // -------------------------------------------------------------------------- // // Default constructor. Set a rule (phrase), see class description for more // details. Set a name and title. If no title is given it is set to the rule. // MDataPhrase::MDataPhrase(const char *rule, const char *name, const char *title) : fFormula(0) { #if ROOT_VERSION_CODE < ROOT_VERSION(6,00,00) #define TFP TFormulaPrimitive #else #define TFP ROOT::v5::TFormulaPrimitive #endif // More in TFormulaPrimitive.cxx // More in TFormulaMathInterface if (!TFP::FindFormula("isnan")) { TFP::AddFormula(new TFP("log2", "log2", (TFP::GenFunc10)TMath::Log2)); TFP::AddFormula(new TFP("fabs", "fabs", (TFP::GenFunc10)TMath::Abs)); TFP::AddFormula(new TFP("floor", "floor", (TFP::GenFunc10)TMath::Floor)); TFP::AddFormula(new TFP("ceil", "ceil", (TFP::GenFunc10)TMath::Ceil)); TFP::AddFormula(new TFP("nint", "nint", (TFP::GenFunc10)MFastFun::Nint)); TFP::AddFormula(new TFP("round", "round", (TFP::GenFunc10)MFastFun::Nint)); TFP::AddFormula(new TFP("sgn", "sgn", (TFP::GenFunc10)MFastFun::Sign)); TFP::AddFormula(new TFP("isnan", "isnan", (TFP::GenFunc10)MFastFun::IsNaN)); TFP::AddFormula(new TFP("finite", "finite", (TFP::GenFunc10)MFastFun::Finite)); } // TFormulaPrimitive is used to get direct acces to the function pointers // GenFunc - pointers to the static function // TFunc - pointers to the data member functions // // The following sufixes are currently used, to describe function arguments: // ------------------------------------------------------------------------ // G - generic layout - pointer to double (arguments), pointer to double (parameters) // 10 - double // 110 - double, double // 1110 - double, double, double fName = name ? name : "MDataPhrase"; fTitle = title ? title : rule; fMembers.SetOwner(); if (rule) SetRule(rule); // Make sure that all object deleted are also deleted from this list gROOT->GetListOfCleanups()->Add(this); // Make sure that this doesn't remain in ListOfCleanups after deletion SetBit(kMustCleanup); } // -------------------------------------------------------------------------- // // Destructor // MDataPhrase::~MDataPhrase() { if (fFormula) delete fFormula; } // -------------------------------------------------------------------------- // // CopyConstructor // MDataPhrase::MDataPhrase(MDataPhrase &ts) { TFormula *f = ts.fFormula; fName = "MDataPhrase"; fTitle = f ? f->GetExpFormula() : (TString)""; fFormula = f ? (TFormula*)f->Clone() : 0; fFormula->SetBit(kMustCleanup); gROOT->GetListOfFunctions()->Remove(fFormula); fMembers.SetOwner(); TObject *o = NULL; TIter Next(&ts.fMembers); while ((o=Next())) fMembers.AddLast(o->Clone()); // Make sure that all object deleted are also deleted from this list gROOT->GetListOfCleanups()->Add(this); // Make sure that this doesn't remain in ListOfCleanups after deletion SetBit(kMustCleanup); } // -------------------------------------------------------------------------- // // Evaluates and returns the result of the member list. // Double_t MDataPhrase::Eval(const Double_t *x) const { const Int_t n = fMembers.GetEntriesFast(); if (fMembers.GetSize()fStorage Double_t *arr = fStorage.GetArray(); // Evaluate parameters (the access with TObjArray::UncheckedAt is // roughly two times faster than with a TIter and rougly three // times than accessing a TOrdCollection. However this might still // be quite harmless compared to the time needed by GetValue) for (Int_t i=0; i(fMembers.UncheckedAt(i))->GetValue(); // Evaluate function return fFormula->EvalPar(x, arr); } // -------------------------------------------------------------------------- // // Returns kTRUE if all members of fMemebers are valid and fFormula!=NULL // Bool_t MDataPhrase::IsValid() const { TIter Next(&fMembers); MData *data = NULL; while ((data=(MData*)Next())) if (!data->IsValid()) return kFALSE; return fFormula ? kTRUE : kFALSE; } // -------------------------------------------------------------------------- // // Checks whether at least one member has the ready-to-save flag. // Bool_t MDataPhrase::IsReadyToSave() const { TIter Next(&fMembers); MData *data = NULL; while ((data=(MData*)Next())) if (data->IsReadyToSave()) return kTRUE; return kFALSE; } // -------------------------------------------------------------------------- // // PreProcesses all members in the list // Bool_t MDataPhrase::PreProcess(const MParList *plist) { if (!fFormula) { *fLog << err << "Error - So far no valid phrase was setup." << endl; return kFALSE; } TIter Next(&fMembers); MData *member=NULL; // // loop over all members // while ((member=(MData*)Next())) if (!member->PreProcess(plist)) { *fLog << err << "Error - Preprocessing Data Member "; *fLog << member->GetName() << " in " << fName << endl; return kFALSE; } // For speed reasons MArrayD instead of TArrayD is used // (no range check is done when accessing). The storage is // allocated (Set) only when its size is changing. If GetValue // is called many times this should improve the speed significantly // because the necessary memory is already allocated and doesn't need // to be freed. (Just a static variable is not enough, because there // may be several independant objects of this class) fStorage.Set(fMembers.GetSize()); return kTRUE; } // -------------------------------------------------------------------------- // // Builds a rule from all the list members. This is a rule which could // be used to rebuild the list using the constructor of a MDataChain // TString MDataPhrase::GetRule() const { if (!fFormula) return ""; TString rule = fFormula->GetTitle(); //fFormula->GetExpFormula(); MData *member = NULL; Int_t i=0; TIter Next(&fMembers); while ((member=(MData*)Next())) { TString r = member->GetRule(); r.ReplaceAll("]", "\\]"); rule.ReplaceAll(Form("[%d]", i++), r); } rule.ReplaceAll("\\]", "]"); return rule; } // -------------------------------------------------------------------------- // // This returns the rule as seen/interpreted by the TFormula. Mainly // for debugging purposes // TString MDataPhrase::GetRuleRaw() const { if (!fFormula) return ""; return fFormula->GetExpFormula(); } // -------------------------------------------------------------------------- // // Return the value converted to a Bool_t // Bool_t MDataPhrase::GetBool() const { return TMath::Nint(GetValue())!=0; } /* // -------------------------------------------------------------------------- // // Return a comma seperated list of all data members used in the chain. // This is mainly used in MTask::AddToBranchList // TString MDataPhrase::GetDataMember() const { TString str; TIter Next(&fMembers); MData *member=(MData*)Next(); if (!member->GetDataMember().IsNull()) str += member->GetDataMember(); while ((member=(MData*)Next())) { if (!member->GetDataMember().IsNull()) { str += ","; str += member->GetDataMember(); } } return str; } */ void MDataPhrase::SetVariables(const TArrayD &arr) { fMembers.R__FOR_EACH(MData, SetVariables)(arr); } // -------------------------------------------------------------------------- // // Check for corresponding entries in resource file and setup data phrase. // // Assuming your MDataChain is called (Set/GetName): MyData // // Now setup the condition, eg: // MyData.Rule: log10(MHillas.fSize) // or // MyData.Rule: log10(MHillas.fSize) - 4.1 // // If you want to use more difficult rules you can split the // condition into subrules. Subrules are identified // by {}-brackets. Avoid trailing 0's! For example: // // MyData.Rule: log10(MHillas.fSize) + {0} - {1} // MyData.0: 5.5*MHillas.fSize // MyData.1: 2.3*log10(MHillas.fSize) // // The numbering must be continous and start with 0. You can use // a subrules more than once. All {}-brackets are simply replaced // by the corresponding conditions. The rules how conditions can // be written can be found in the class description of MDataChain. // Int_t MDataPhrase::ReadEnv(const TEnv &env, TString prefix, Bool_t print) { Bool_t rc = kFALSE; if (!IsEnvDefined(env, prefix, "Rule", print)) return rc; TString rule = GetEnvValue(env, prefix, "Rule", ""); Int_t idx=0; while (1) { TString cond; if (IsEnvDefined(env, prefix, Form("%d", idx), print)) { cond += "("; cond += GetEnvValue(env, prefix, Form("%d", idx), ""); cond += ")"; } if (cond.IsNull()) break; rule.ReplaceAll(Form("{%d}", idx), cond); idx++; } if (rule.IsNull()) { *fLog << warn << "MDataPhrase::ReadEnv - ERROR: Empty rule found." << endl; return kERROR; } if (!SetRule(rule)) return kERROR; if (print) *fLog << inf << "found: " << GetRule() << endl; return kTRUE; }