sewing_hasher.cc

Go to the documentation of this file.

// -*- mode:c++;tab-width:2;indent-tabs-mode:t;show-trailing-whitespace:t;rm-trailing-spaces:t -*-
// vi: set ts=2 noet;
//
// (c) Copyright Rosetta Commons Member Institutions.
// (c) This file is part of the Rosetta software suite and is made available under license.
// (c) The Rosetta software is developed by the contributing members of the Rosetta Commons.
// (c) For more information, see http://www.rosettacommons.org. Questions about this can be
// (c) addressed to University of Washington UW TechTransfer, email: license@u.washington.edu.

/// @file Apps/sewing_hasher.cc
///
/// @brief An MPI-enabled application that reads in a Hasher hashtable, scores each model in that table against all others, and
/// generates a SewGraph file.
///
/// @author Tim Jacobs

//Package headers
#include <devel/init.hh>
#include <protocols/sewing/hashing/Hasher.hh>
#include <protocols/sewing/conformation/Model.hh>
#include <protocols/sewing/util/io.hh>

//Protocol headers
#include <core/types.hh>
#include <core/id/AtomID.hh>
#include <core/pose/Pose.hh>
#include <core/import_pose/import_pose.hh>

//Utility headers
#include <utility/io/izstream.hh>
#include <utility/mpi_util.hh>
#include <utility/string_util.hh>
#include <utility/vector1.hh>
#include <utility/sql_database/DatabaseSessionManager.hh>
#include <utility/file/FileName.hh>

#include <basic/options/option.hh>
#include <basic/options/keys/sewing.OptionKeys.gen.hh>
#include <basic/options/keys/in.OptionKeys.gen.hh>
#include <basic/options/keys/inout.OptionKeys.gen.hh>
#include <basic/Tracer.hh>
#include <basic/database/sql_utils.hh>

//C++ headers
#include <map>
#include <set>
#include <iterator>

static basic::Tracer TR("sewing_hasher");

int
main( int argc, char * argv [] ) {
  using namespace basic::options;
  using namespace basic::options::OptionKeys;
  using namespace protocols::sewing;

  try {
    // initialize core and read options
    devel::init(argc, argv);

    if ( !option[sewing::mode].user() ) {
      std::stringstream err;
      err << "You must provide a mode for sewing_hasher to run in using the -sewing:mode flag. Valid options are" << std::endl;
      err << "  -generate: generates a model file from an sqlite database" << std::endl;
      err << "  -generate_five_ss_model: generates a 3~5 ss model file from an sqlite database" << std::endl;
      err << "  -hash: score all models against each other and create a plain text score file (MPI required)" << std::endl;
      err << "  -convert: convert a plain text score file to a binary score file. This is required by the SEWING movers" << std::endl;
      utility_exit_with_message(err.str());
    }

    //Check for model file (either for reading or writing)
    std::map< int, Model > models;
    std::string model_filename = option[sewing::model_file_name];
    if ( !option[sewing::model_file_name].user() ) {
      std::stringstream err;
      err << "You must provide a model file name to the sewing_hasher using the -model_file_name flag. To generate a new model file use the "
        << "-generate_models_from_db flag and a new model file with that name will be written. Otherwise, the model file will be read";
      utility_exit_with_message(err.str());
    }

    //////////////////////////// MODEL GENERATION ///////////////////////////////////

    //Are we just generating a model file?
    if ( option[sewing::mode].value() == "generate" ) {

      //Create comments stream for model file and add the date
      std::stringstream comments;
      //timestamp became commented by Vikram to prevent daily integration failure
      //time_t t = time(0); // get time now
      //struct tm * now = localtime( & t );
      //comments << "#Model file created on " << (now->tm_year + 1900) << '-'
      // << (now->tm_mon + 1) << '-'
      // << now->tm_mday
      // << std::endl;

      bool hash_tag_only_terminal_Es = option[sewing::hash_tag_only_terminal_Es].def(false);
      TR << "hash_tag_only_terminal_Es: " << hash_tag_only_terminal_Es << std::endl;
      std::string hash_between;
      //std::string model_three_ss_filename;

      //Generate model file from a list of PDBs. All models will have 1 segment
      if ( option[ in::file::l ].user() ) {
        comments << "#Models generated from PDB input (-l flag)" << std::endl;
        utility::vector1<utility::file::FileName> input_lists( option[ in::file::l ]() );
        utility::vector1<utility::file::FileName> pdb_library;
        for ( core::Size i = 1; i <= input_lists.size(); i++ ) {
          utility::io::izstream current_input_list( input_lists[i] );
          if ( !current_input_list.good() ) {
            utility_exit_with_message("unable to open input file file: "+input_lists[i].name()+"\n");
          }
          while ( current_input_list.good() ) {
            std::string name;
            current_input_list.getline(name);
            if ( current_input_list.good() ) pdb_library.push_back( utility::file::FileName(name) );
          }
        }

        for ( core::Size i=1; i<=pdb_library.size(); ++i ) {
          core::pose::Pose pose;
          core::import_pose::pose_from_file(pose, pdb_library[i], core::import_pose::PDB_file);
          utility::vector1< std::pair<core::Size,core::Size> > segments;
          segments.push_back(std::make_pair(1, pose.total_residue()));
          Model pdb_model = create_model_from_pose(pose, segments, (int)i);
          models.insert(std::make_pair(i, pdb_model));
        }

      } else {
        //Generate models from a features database. Each segment is a single piece of secondary structure

        if ( hash_tag_only_terminal_Es ) {
          hash_between = "hash_tag_only_terminal_Es";
          comments << "#Only terminal Es are hash bool true to be merged with other nodes later (but as of 2015/11/14, later model assembly has 'ERROR: alignment_scores.size() == 1'" << std::endl;
          // model_three_ss_filename = model_filename + "_three_ss_will_be_hashed_only_between_Es";
          //(not possible due to integration test)
        } else {
          hash_between = "hash_between_any_HEs";
          comments << "# hash between any HEs are bool true to be merged with other nodes" << std::endl;
          // model_three_ss_filename = model_filename + "_three_ss_will_be_hashed_between_any_HEs";
          //(not possible due to integration test)
        }

        if ( !option[ sewing::assembly_type ].user() ) {
          std::stringstream err;
          err << "You must provide an assembly_type (continuous or discontinuous) with the -sewing:assembly_type flag in order to extract models";
          utility_exit_with_message(err.str());
        }
        if ( option[ sewing::assembly_type ].value() == "discontinuous" ) {
          comments << "#Discontinuous models generated from sqlite database " << option[inout::dbms::database_name].value() << std::endl;
          models = get_discontinuous_models_from_db();
        } else if ( option[ sewing::assembly_type ].value() == "continuous" ) {
          comments << "#Continuous models generated from sqlite database " << option[inout::dbms::database_name].value() << std::endl;
          models = get_continuous_models_from_db(hash_between);
        }
      }

      write_model_file(comments.str(), models, model_filename);
      TR << "New model file " << model_filename << " successfully written." << std::endl;
      std::exit(0);
    } else if ( option[sewing::mode].value() == "generate_five_ss_model" ) {

      //Create comments stream for model file and add the date
      std::stringstream comments;
      //timestamp became commented by Vikram to prevent daily integration failure
      //time_t t = time(0);   // get time now
      //struct tm * now = localtime( & t );
      //comments << "#Model file created on " << (now->tm_year + 1900) << '-'
      // << (now->tm_mon + 1) << '-'
      // <<  now->tm_mday
      // << std::endl;

      //Generate models from a features database. Each segment is a single piece of secondary structure
      comments << "# 3 or 5 secondary structures based models generated from sqlite database " << option[inout::dbms::database_name].value() << std::endl;

      bool hash_tag_only_terminal_Es = option[sewing::hash_tag_only_terminal_Es].def(false);
      TR << "hash_tag_only_terminal_Es: " << hash_tag_only_terminal_Es << std::endl;
      std::string hash_between;
      std::string model_five_ss_filename;
      if ( hash_tag_only_terminal_Es ) {
        hash_between = "hash_tag_only_terminal_Es";
        comments << "# only_terminal_Es_are_hashed " << std::endl;
        model_five_ss_filename = model_filename + "_three_or_five_ss_will_be_hashed_only_between_Es";
      } else {
        hash_between = "hash_between_any_HEs";
        comments << "# will_be_merged_between_any_HEs " << std::endl;
        model_five_ss_filename = model_filename + "_three_or_five_ss_will_be_hashed_between_any_HEs";
      }

      std::map< int, Model > models = get_5_ss_models_from_db(hash_between);

      write_model_file(comments.str(), models, model_five_ss_filename);
      TR << "New model file with 3~5 ss " << model_five_ss_filename << " successfully written." << std::endl;

      std::exit(0);
    }

    //If we aren't generating models, then we need to read them
    models = read_model_file(model_filename);

    ///////////////////////// BINARY FILE TESTING //////////////////////////////

    if ( option[sewing::mode].value() == "test" ) {
      std::string binary_filename = option[sewing::score_file_name].value();
      SewGraphOP graph( new SewGraph(models, 1) );
      graph->report_binary_stats(models, binary_filename);
    }

    ///////////////////////// MODEL CONVERSION TO BINARY //////////////////////////////

    //If we are generating a binary file then do that and exit
    if ( ( option[sewing::mode].value() == "convert" ) ) {
      if ( !option[sewing::score_file_name].user() ) {
        std::stringstream err;
        err << "You must provide a score file name to the sewing_hasher for binary conversion.";
        utility_exit_with_message(err.str());

      }
      std::string binary_filename = utility::to_string(option[sewing::score_file_name].value())+utility::to_string(".bin");

      SewGraphOP graph;
      if ( option[ sewing::assembly_type ].value() == "discontinuous" ) {
        graph = SewGraphOP( new SewGraph(models, 2) );
      } else if ( option[ sewing::assembly_type ].value() == "continuous" ) {
        graph = SewGraphOP( new SewGraph(models, 1) );
      }
      graph->generate_binary_score_file(option[sewing::score_file_name].value(), binary_filename);
      std::exit(0);
    }


    //////////// MODEL COMPARISON USING GEOMETRIC HASHING /////////////////

    if ( ( option[sewing::mode].value() == "hash" ) ) {
      if ( !option[sewing::score_file_name].user() ) {
        utility_exit_with_message("You must provide a graph file name to the sewing_hasher using -score_file_name");
      }
      core::Size min_hash_score = option[sewing::min_hash_score].def(10);
      core::Size max_clash_score = option[sewing::max_clash_score].def(0);
      core::Size num_segments_to_match = option[sewing::num_segments_to_match].def(1);
      core::Size box_length = option[sewing::box_length].def(3);
      // box_length 3 is for neighborhood lookup box size 27 (=3^3), while box_length 5 is for neighborhood lookup box size 125 (=5^3)

      std::string score_file_name = option[sewing::score_file_name];


      ///// Size to string reference: http://www.cplusplus.com/articles/D9j2Nwbp/
      std::ostringstream convert_min_hash_score;
      std::string min_hash_score_string;
      convert_min_hash_score << min_hash_score;
      min_hash_score_string = convert_min_hash_score.str();

      std::ostringstream convert_max_clash_score; // using single 'std::ostringstream convert' many times stopped to initialize convert, so I begin to use many 'std::ostringstream convert' since 2016/1/19
      std::string max_clash_score_string;
      convert_max_clash_score << max_clash_score;
      max_clash_score_string = convert_max_clash_score.str();

      std::ostringstream convert_num_segments_to_match;
      std::string num_segments_to_match_string;
      convert_num_segments_to_match << num_segments_to_match;
      num_segments_to_match_string = convert_num_segments_to_match.str();

      std::ostringstream convert_box_length;
      std::string box_length_string;
      convert_box_length << box_length;
      box_length_string = convert_box_length.str();

      /*
      OK with release compilation, but not ok with mac clang debug
      "src/apps/public/sewing/sewing_hasher.cc:249:74: error: cannot take the address of an rvalue of type 'std::__1::basic_ostringstream<char, std::__1::char_traits<char>, std::__1::allocator<char> >'
      std::string min_hash_score_string = static_cast<std::ostringstream*>( &(std::ostringstream() << min_hash_score) )->str();
      ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"

      std::string min_hash_score_string = static_cast<std::ostringstream*>( &(std::ostringstream() << min_hash_score) )->str();
      std::string max_clash_score_string = static_cast<std::ostringstream*>( &(std::ostringstream() << max_clash_score) )->str();
      std::string num_segments_to_match_string = static_cast<std::ostringstream*>( &(std::ostringstream() << num_segments_to_match) )->str();
      std::string box_length_string = static_cast<std::ostringstream*>( &(std::ostringstream() << box_length) )->str();
      */

      std::string hash_target = option[sewing::mode].value();


      // Keep this comment!  Customized_score_file_name is very useful, Doonam just commented here to avoid integration test failure
      // score_file_name = score_file_name + "_" + hash_target + "_" + min_hash_score_string + "_min_hash_score_" + max_clash_score_string + "_max_clash_score_" + num_segments_to_match_string + "_num_segments_to_match_" + box_length_string + "_box_length";
      ///////////////////


      TR << "Bundle Hasher options:" << std::endl;
      TR << "\tMinimum Score: " << min_hash_score << std::endl;
      TR << "\tMaximum Clash Score: " << max_clash_score << std::endl;
      TR << "\tNumber of segments to match: " << num_segments_to_match << std::endl;
      TR << "\tScore file name: " << score_file_name << std::endl;
      TR << "\tNeighborhood lookup box_length: " << box_length << std::endl;

#ifdef USEMPI
        //Now, score models. Split this work between multiple processors if we have them
        core::Size n_models;
        if ( option[sewing::max_models].user() ) {
          n_models = std::min( (core::Size)option[sewing::max_models], models.size() );
        } else {
          n_models = models.size();
        }

        int starting_model = 0;
        if ( option[sewing::starting_model].user() ) {
          starting_model = option[sewing::starting_model].value();
        }

        core::Size rank = utility::mpi_rank();
        if(rank == 0) {
          core::Size num_procs = utility::mpi_nprocs();

          TR << "Master node has " << n_models << " jobs to submit" << std::endl;

          //send out initial jobs
          std::map< int, Model >::const_iterator it = models.begin();
          ++it; //Increment the iterator (no need to score the first model against just itself)

          std::map< int, Model >::const_iterator it_end = models.begin();
          std::advance(it_end, n_models);

          if(num_procs > n_models) {
            utility_exit_with_message("You have more processors than number of models to hash. Reduce your number of processors for efficiency");
          }

          core::Size curr_proc = 1;
          for(; it != it_end; ++it) {
            //send *it to curr_proc
            if(it->first >= starting_model) {
              utility::send_integer_to_node(curr_proc,it->first);
              TR << "Master node sent a job for model " << it->first << " to processor " << curr_proc << std::endl;
              ++curr_proc;
              if(curr_proc == num_procs) { break; }
            }
          }

          ++it;//Increment iterator once more to account for the last model sent in the above loop
          while( it != it_end ){
            //wait for a processor to send you a message
            //send more work
            core::Size received_node = utility::receive_integer_from_anyone();
            TR << "Master node received a message from processor " << received_node << std::endl;
            utility::send_integer_to_node(received_node,it->first);
            TR << "Master node sent a new job for model " << it->first << " to processor " << received_node << std::endl;
            ++it;
          }

          //Once we're done with all the work, tell each processor to spin down
          core::Size counter = 1;
          while(counter != num_procs){
            core::Size received_node = utility::receive_integer_from_anyone();
            utility::send_integer_to_node(received_node,0);
            ++counter;
          }
          TR << "Master node finished sending jobs" << std::endl;
        }

        else {
          while(true) {
            //Recieve message from master node
            int model_id = utility::receive_integer_from_node(0);
            if(model_id == 0){ break; }

            TR << "Processor " << rank << " received job for model number " << model_id << std::endl;

            //Breakup the models into groups of 1000 for scoring. This make the memory demand *much* less
            //with a minor hit to scoring speed.
            std::map< int, Model >::const_iterator it = models.begin();
            std::map< int, Model >::const_iterator it_end = models.find(model_id);
            while(true) {
              core::Size counter = 0;
              Hasher hasher;
              for(; it != it_end; ++it) {
                hasher.insert(it->second);
                ++counter;
                if(counter > 100) {
                  ++it;
                  break;
                }
              }
            ScoreResults group_scores = hasher.score(models[model_id], num_segments_to_match, min_hash_score, max_clash_score, false, box_length);
              hasher.remove_connection_inconsistencies(models, group_scores);
              std::string node_score_file_name = score_file_name + "." + utility::to_string(rank);
              write_hashing_scores_to_file(group_scores, node_score_file_name);
              TR << "Processor " << rank << " done scoring. Found " << group_scores.size() << " valid comparisons" << std::endl;
              if(it == it_end) {
                break;
              }
            }

//            //Begin work
//            Hasher hasher;
//            ScoreResults scores;
//            for(; it != it_end; ++it) {
//              if(it->first == model_id) {
//                TR << "Processor " << rank << " begin scoring " << model_id << std::endl;
//                scores = hasher.score(it->second, 1, min_hash_score, 0, false);
//                break;
//              }
//              hasher.insert(it->second);
//            }
//            hasher.remove_connection_inconsistencies(models, scores);

            //std::string node_score_file_name = score_file_name + "." + utility::to_string(rank);
            //write_hashing_scores_to_file(group_scores, node_score_file_name);

            //Send message back to master node
            utility::send_integer_to_node(0, rank);
      }
        TR << "Processor " << rank << " was told to stop working" << std::endl;
    }
        MPI_Barrier( MPI_COMM_WORLD ); // make all nodes reach this point together.
        MPI_Finalize();
#else // so use 1 cpu
      Hasher hasher;
      std::map< int, Model >::const_iterator it1 = models.begin();
      std::map< int, Model >::const_iterator it_end = models.end();
      std::map< int, Model >::const_iterator it2 = models.begin();
      for ( ; it1 != it_end; ++it1 ) {
        ScoreResults scores;
        for ( ; it2 != it1; ++it2 ) {
          hasher.insert(it2->second); // it2->second is Model itself
        }
        TR << "current model id (it1): " << (it1->second).model_id_ << std::endl;
        TR << "current model id (it2): " << (it2->second).model_id_ << std::endl;
        scores = hasher.score(it1->second, num_segments_to_match, min_hash_score, max_clash_score, true, box_length);
        if ( ! option[sewing::do_not_remove_connection_inconsistencies].user() ) {
          option[sewing::do_not_remove_connection_inconsistencies].value( 0 );
        }
        bool do_not_remove_connection_inconsistencies = option[sewing::do_not_remove_connection_inconsistencies];
        TR << "do_not_remove_connection_inconsistencies: " << do_not_remove_connection_inconsistencies << std::endl;
        if ( (!do_not_remove_connection_inconsistencies) ) {
          //remove edges between segments that both have 'next' or 'previous' segments
          hasher.remove_connection_inconsistencies(models, scores);
        }
        TR << "Done scoring the " << it2->first << "th model (this is not necessariliy a model_id) found (" << scores.size() << ") valid comparisons" << std::endl;
        if ( scores.size() > 0 && TR.Debug ) {
          BasisPair bp = scores.begin()->first;
          std::map< SegmentPair, core::Size > segment_matches = scores.begin()->second.segment_match_counts;
          std::map< SegmentPair, AtomMap > segment_matches2 = scores.begin()->second.segment_matches;
          TR.Debug << "After scoring." << std::endl;
          TR.Debug << "\tModels: " << bp.first.model_id << " " << bp.second.model_id << std::endl;
          TR.Debug << "\tBasis Residues: " << bp.first.resnum << " " << bp.second.resnum << std::endl;
          TR.Debug << "\tNumber of matched segments: " << segment_matches.size() << std::endl;
          std::map< SegmentPair, core::Size >::const_iterator it = segment_matches.begin();
          std::map< SegmentPair, core::Size >::const_iterator it_end = segment_matches.end();
          std::map< SegmentPair, AtomMap >::const_iterator it2 = segment_matches2.begin();
          for ( ; it != it_end; ++it ) {
            TR.Debug << "\tSegments " << it->first.first << " and " << it->first.second << " have " << it->second << " overlapping atoms." << std::endl;
            TR.Debug << "\tSegments " << it2->first.first << " and " << it2->first.second << " have " << it2->second.size() << " overlapping atoms." << std::endl;
            ++it2;
          }
        }
        write_hashing_scores_to_file(scores, score_file_name);
      } //for ( ; it1 != it_end; ++it1 ) {
#endif
    }// if ( ( option[sewing::mode].value() == "hash" ) )
  } catch ( utility::excn::EXCN_Base& excn ) {
    std::cerr << "Exception : " << std::endl;
    excn.show( std::cerr );
    return -1;
  }
  return 0;
}