// // iqtreemixhmm.h // tree // // Created by Thomas Wong on 19/01/23. // #ifndef iqtreemixhmm_h #define iqtreemixhmm_h #include #include "iqtreemix.h" #include "tree/phylohmm.h" #include "model/modelhmm.h" #include "model/modelhmmgm.h" #include "model/modelhmmtm.h" class IQTreeMixHmm : public IQTreeMix, public PhyloHmm { public: /** default constructor */ IQTreeMixHmm(); IQTreeMixHmm(Params ¶ms, Alignment *aln, vector &trees); /** default destructor */ ~IQTreeMixHmm(); // initialize the model void initializeModel(Params ¶ms, string model_name, ModelsBlock *models_block); // initialize the transition model void initializeTransitModel(Params ¶ms); // initialize the parameters void initializeParams(); // set the tree weights according to the marginal probabilities along the sites void setWeightToMarginalProb(); // compute the log of dotproduct of the logorithm arrays double logDotProd(double* ln_x, double* ln_y, int n); // obtain the log-likelihoods for every pattern and every tree // output ptn_like_cat[i * ntree + j] : log-likelihood of pattern i and tree j void computeLogLikelihoodSiteTree(int updateTree = -1); // compute backward log-likelihood virtual double computeLikelihood(double *pattern_lh = NULL, bool save_log_value = true); /** optimize all branch lengths of one tree @param iterations number of iterations to loop through all branches */ void optimizeAllBranchesOneTree(int whichtree, int my_iterations = 100, double tolerance = TOL_LIKELIHOOD, int maxNRStep = 100); /** optimize all branch lengths of all trees @param iterations number of iterations to loop through all branches @return the likelihood of the tree */ double optimizeAllBranches(double* pattern_mix_lh = NULL, int my_iterations = 100, double tolerance = TOL_LIKELIHOOD, int maxNRStep = 100); double optimizeAllBranchLensByBFGS(double gradient_epsilon, double logl_epsilon, int maxsteps = 3); /** @return true if this is a HMM model */ virtual bool isHMM() { return true; } virtual void startCheckpoint(); virtual string optimizeModelParameters(bool printInfo, double logl_epsilon); // Optimize parameters according to the MAST model string optimizeModelParamMAST(bool printInfo, double logl_epsilon); // Optimize parameters according to the HMM model string optimizeModelParamHMM(bool printInfo, double logl_epsilon); virtual void setNumThreads(int num_threads); /** test the best number of threads */ virtual int testNumThreads(); virtual int getNParameters(); // print out all the results to a file void printResults(const char *filename, int cat_assign_method = 0); // print out the marginal probabilities to a file void printMarginalProb(const char *filename); // show the values of the parameters void showParameters(ostream& out); // optimize all substitution models double optimizeAllSubstModels(double gradient_epsilon, double* pattern_mix_lh = NULL); // optimize all RHAS models double optimizeAllRHASModels(double gradient_epsilon, double score = 0.0, double* pattern_mix_lh = NULL); private: // indicate which tree's unlinked parameters is under optimization // -1 : parameters affecting all trees (default) // >= 0 : parameters affecting a specific tree int optimTree; // indicate which branch group is under optimization // -1 : no branch group is under optimization // >=0 : optimizing a specific branch group int optimBranchGrp; // which objective function (default: MAST) // 0: backLikelihood // 1: MAST int objFun; // whether using the optimization engine in IQTreeMix bool isTMixOptimEngine; string* objAlgo; // branch lengths of all the trees vector allbranchlens; // type of all sites // 0 - parsimony informatic; 1 - invariant (including constant and e.g. GS--G-GGG (S = G/C)); // 2 - uninformatic but not invariant (e.g. GTTTTTT) int* siteTypes; // get the type of all sites for type-dependent HMM model // 0 - parsimony informatic; 1 - invariant (including constant and e.g. GS--G-GGG (S = G/C)); // 2 - uninformatic but not invariant (e.g. GTTTTTT) void setSiteTypes(); // compute the log-likelihoods for a single tree t void computeLogLikelihoodSingleTree(int t); // get the branch lengths of all trees to the variable allbranchlens void getAllBranchLengths(); // set the branch lengths of all trees from the variable allbranchlens void setAllBranchLengths(); // show the branch lengths of all trees void showAllBranchLengths(); //-------------------------------------------- // optimization of branch lengths using BFGS //-------------------------------------------- // the following three functions are for dimension = 1 double computeFunction(double x); double setSingleVariable(); void getSingleVariable(double x); // the following four functions are for dimension > 1 virtual double targetFunk(double x[]); virtual void setVariables(double *variables); virtual void getVariables(double *variables); virtual void setBounds(double *lower_bound, double *upper_bound, bool *bound_check); virtual int getNDim(); double optimizeBranchGroup(int branchgrp, double gradient_epsilon); void showBranchGrp(); /** If there are multiple branches belonging to the same group set all the branches of the same group to their average */ void setAvgLenEachBranchGrp(); // update the ptn_freq array according to the marginal probabilities along each site for each tree void computeFreqArray(double* pattern_mix_lh = NULL, bool need_computeLike = true, int update_which_tree = -1); // get marginal probabilities along each site for each tree void getMarginalProb(bool need_computeLike = true, int update_which_tree = -1); }; #endif /* iqtreemixhmm_h */