X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FText%2FTradition%2FAnalysis.pm;h=4842b69261fbb2cc0266eb0b3ee9018250520e95;hb=3837c155d39333869a93adf1e8375960ffbf3a92;hp=a64277ffd7f57bafeebad7c41ac75757c1f58bbe;hpb=c4a4fb1ba91d074f14540bdd11ca0caf27bf56c4;p=scpubgit%2Fstemmatology.git diff --git a/lib/Text/Tradition/Analysis.pm b/lib/Text/Tradition/Analysis.pm index a64277f..4842b69 100644 --- a/lib/Text/Tradition/Analysis.pm +++ b/lib/Text/Tradition/Analysis.pm @@ -2,96 +2,38 @@ package Text::Tradition::Analysis; use strict; use warnings; +use Benchmark; +use Exporter 'import'; use Text::Tradition; use Text::Tradition::Stemma; -sub new { - my( $class, $args ) = @_; - my $self = {}; - bless( $self, $class ); - $self->{'data'} = []; - foreach my $t ( @{$args->{'traditions'}} ) { - $self->run_analysis( $t->{'file'}, $t->{'stemmadot'} ); - } - return $self; -} +use vars qw/ @EXPORT_OK /; +@EXPORT_OK = qw/ run_analysis group_variants wit_stringify /; sub run_analysis { - my( $self, $file, $stemmadot ) = @_; + my( $tradition, $stemma ) = @_; # What we will return - my $svg; my $variants = []; my $data = {}; - - # Read in the file and stemma - my $tradition = Text::Tradition->new( - 'input' => 'Self', - 'file' => $file, - 'linear' => 1, - ); - $data->{'title'} = $tradition->name; - - my $stemma = Text::Tradition::Stemma->new( - 'collation' => $tradition->collation, - 'dot' => $stemmadot, - ); - # We will return the stemma picture - $svg = $stemma->as_svg( { size => "8,7.5" } );; - $data->{'svg'} = $svg; - + # We have the collation, so get the alignment table with witnesses in rows. # Also return the reading objects in the table, rather than just the words. my $wits = {}; map { $wits->{$_} = 1 } $stemma->witnesses; - my $all_wits_table = $tradition->collation->make_alignment_table( 'refs', $wits ); - # For each column in the alignment table, we want to see if the existing - # groupings of witnesses match our stemma hypothesis. We also want, at the - # end, to produce an HTML table with all the variants. - my $html_columns = 0; + # groupings of witnesses match our stemma hypothesis. We also need to keep + # track of the maximum number of variants at any one location. + my $max_variants = 0; my ( $total, $genealogical, $conflicts ) = ( 0, 0, 0 ); - # Strip the list of sigla and save it for correlation to the readings. - my $col_wits = shift @$all_wits_table; - - # We will return a data structure, an array for each row that looks like: - # { id = X, genealogical = Y, readings = [ text = X, group = Y], empty = N } - foreach my $i ( 0 .. $#$all_wits_table ) { - # For each column in the table, group the readings by witness. - my $rdg_wits = {}; - my $col_rdgs = shift @$all_wits_table; - my $rank; - my $lacunose = []; - foreach my $j ( 0 .. $#{$col_rdgs} ) { - my $rdg = $col_rdgs->[$j]; - my $rdg_text = '(omitted)'; # Initialize in case of empty reading - if( $rdg ) { - if( $rdg->is_lacuna ) { - $rdg_text = undef; # Don't count lacunae - push( @$lacunose, $col_wits->[$j] ); - } else { - $rdg_text = $rdg->text; - # Get the rank from any real reading; they should be identical. - $rank = $rdg->rank; - } - } - if( defined $rdg_text ) { - # Initialize the witness array if we haven't got one yet - $rdg_wits->{$rdg_text} = [] unless $rdg_wits->{$rdg_text}; - # Add the relevant witness, subject to a.c. logic - add_variant_wit( $rdg_wits->{$rdg_text}, $col_wits->[$j], - $tradition->collation->ac_label ); - } - } - - # See if this column has any potentially genealogical variants. - # If not, skip to the next. - $total++ unless scalar keys %$rdg_wits == 1; - my( $groups, $readings ) = useful_variant( $rdg_wits ); - next unless $groups && $readings; + my $t0 = Benchmark->new(); + my $variant_groups = group_variants( $tradition->collation, $wits ); + foreach my $rank ( 0 .. $#{$variant_groups} ) { + my $groups = $variant_groups->[$rank]->{'groups'}; + my $readings = $variant_groups->[$rank]->{'readings'}; + my $lacunose = $variant_groups->[$rank]->{'lacunose'}; - # Keep track of our widest row - $html_columns = scalar @$groups if scalar @$groups > $html_columns; + $max_variants = scalar @$groups if scalar @$groups > $max_variants; # We can already look up witnesses for a reading; we also want to look # up readings for a given witness. @@ -102,12 +44,11 @@ sub run_analysis { # For all the groups with more than one member, collect the list of all # contiguous vertices needed to connect them. - $DB::single = 1 if $rank == 733; - my $variant_row = analyze_variant_location( $group_readings, $groups, + my $variant_loc = analyze_variant_location( $group_readings, $groups, $stemma->graph, $lacunose ); - $variant_row->{'id'} = $rank; - $genealogical++ if $variant_row->{'genealogical'}; - $conflicts += grep { $_->{'conflict'} } @{$variant_row->{'readings'}}; + $variant_loc->{'id'} = $rank; + $genealogical++ if $variant_loc->{'genealogical'}; + $conflicts += grep { $_->{'conflict'} } @{$variant_loc->{'readings'}}; # Now run the same analysis given the calculated distance tree(s). # my @trees = @{$stemma->distance_trees}; @@ -122,45 +63,99 @@ sub run_analysis { # } # Record that we used this variant in an analysis - push( @$variants, $variant_row ); + push( @$variants, $variant_loc ); } + my $t1 = Benchmark->new(); + print STDERR "Analysis of graph for " . $tradition->name . " took " + . timestr( timediff( $t1, $t0 ) ) . "seconds\n"; - # Go through our variant rows, after we have seen all of them once, + # Go through our variant locations, after we have seen all of them once, # and add the number of empty columns needed by each. foreach my $row ( @$variants ) { - my $empty = $html_columns - scalar @{$row->{'readings'}}; + my $empty = $max_variants - scalar @{$row->{'readings'}}; $row->{'empty'} = $empty; } - # Populate self with our analysis data. $data->{'variants'} = $variants; $data->{'variant_count'} = $total; $data->{'conflict_count'} = $conflicts; $data->{'genealogical_count'} = $genealogical; - push( @{$self->{'data'}}, $data ); + return $data; +} + +sub group_variants { + my( $c, $wits ) = @_; + my $variant_groups = []; + + # We have the collation, so get the alignment table with witnesses in rows. + # Also return the reading objects in the table, rather than just the words. + my $all_wits_table = $c->make_alignment_table( 'refs', $wits ); + # Strip the list of sigla and save it for correlation to the readings. + my $col_wits = shift @$all_wits_table; + # Any witness in the stemma that has no row should be noted. + foreach ( @$col_wits ) { + $wits->{$_}++; # Witnesses present in table and stemma now have value 2. + } + my @not_collated = grep { $wits->{$_} == 1 } keys %$wits; + foreach my $i ( 0 .. $#$all_wits_table ) { + # For each column in the table, group the readings by witness. + my $rdg_wits = {}; + my $col_rdgs = shift @$all_wits_table; + my $lacunose = [ @not_collated ]; + foreach my $j ( 0 .. $#{$col_rdgs} ) { + my $rdg = $col_rdgs->[$j]; + my $rdg_text = '(omitted)'; # Initialize in case of empty reading + if( $rdg ) { + if( $rdg->is_lacuna ) { + $rdg_text = undef; # Don't count lacunae + push( @$lacunose, $col_wits->[$j] ); + } else { + $rdg_text = $rdg->text; + } + } + if( defined $rdg_text ) { + # Initialize the witness array if we haven't got one yet + $rdg_wits->{$rdg_text} = [] unless $rdg_wits->{$rdg_text}; + # Add the relevant witness, subject to a.c. logic + add_variant_wit( $rdg_wits->{$rdg_text}, $col_wits->[$j], + $c->ac_label ); + } + } + + # See if this column has any potentially genealogical variants. + # If not, skip to the next. + my( $groups, $readings ) = useful_variant( $rdg_wits ); + next unless $groups && $readings; + + push( @$variant_groups, + { 'groups' => $groups, 'readings' => $readings, 'lacunose' => $lacunose } ); + } + return $variant_groups; } + + # variant_row -> genealogical # -> readings [ { text, group, conflict, missing } ] sub analyze_variant_location { my( $group_readings, $groups, $graph, $lacunose, $undirected ) = @_; - my %contig; - my %subgraph; + my $contig = {}; + my $subgraph = {}; my $is_conflicted; my $conflict = {}; - my %missing; - map { $missing{$_} = 1 } @$lacunose; + my $missing = {}; + map { $missing->{$_} = 1 } @$lacunose; my $variant_row = { 'readings' => [] }; # Mark each ms as in its own group, first. foreach my $g ( @$groups ) { my $gst = wit_stringify( $g ); - map { $contig{$_} = $gst } @$g; + map { $contig->{$_} = $gst } @$g; } # Now for each unmarked node in the graph, initialize an array # for possible group memberships. We will use this later to # resolve potential conflicts. - map { $contig{$_} = [] unless $contig{$_} } $graph->vertices; + map { $contig->{$_} = [] unless $contig->{$_} } $graph->vertices; foreach my $g ( sort { scalar @$b <=> scalar @$a } @$groups ) { my $gst = wit_stringify( $g ); # This is the group name my $reachable = { $g->[0] => 1 }; @@ -168,7 +163,7 @@ sub analyze_variant_location { my $part = $graph->copy; my $group_root; $part->delete_vertices( - grep { !ref( $contig{$_} ) && $contig{$_} ne $gst } $graph->vertices ); + grep { !ref( $contig->{$_} ) && $contig->{$_} ne $gst } $graph->vertices ); # Now look to see if our group is connected. if( $undirected ) { # For use with distance trees etc. @@ -183,7 +178,7 @@ sub analyze_variant_location { # Dispense with the trivial case of one reading. # We have to take directionality into account. # How many root nodes do we have? - my @roots = grep { ref( $contig{$_} ) || $contig{$_} eq $gst } + my @roots = grep { ref( $contig->{$_} ) || $contig->{$_} eq $gst } $part->source_vertices; # Assuming that @$g > 1, find the first root node that has at # least one successor belonging to our group. If this reading @@ -191,19 +186,14 @@ sub analyze_variant_location { # that implicitly later. my $nodes_in_subtree = 0; foreach my $root ( @roots ) { - # Prune the tree down to the first root that either is extant - # or has more than one successor. - while( $part->successors( $root ) == 1 && ref $contig{$root} ) { - my @nextroot = $part->successors( $root ); - $part->delete_vertex( $root ); - $root = $nextroot[0]; - } + # Prune the tree to get rid of extraneous hypotheticals. + $root = prune_subtree( $part, $root, $contig ); # Get all the successor nodes of our root. my $tmp_reach = { $root => 1 }; map { $tmp_reach->{$_} = 1 } $part->all_successors( $root ); # Skip this root if none of our successors are in our group # (e.g. isolated 'hypothetical' witnesses with no group) - next unless grep { $contig{$_} } keys %$tmp_reach; + next unless grep { $contig->{$_} } keys %$tmp_reach; if( keys %$tmp_reach > $nodes_in_subtree ) { $nodes_in_subtree = keys %$tmp_reach; $reachable = $tmp_reach; @@ -217,16 +207,16 @@ sub analyze_variant_location { # group. Paint the 'hypotheticals' with our group while we are at it, # unless there is a conflict present. foreach ( keys %$reachable ) { - if( ref $contig{$_} ) { - push( @{$contig{$_}}, $gst ); - } elsif( $contig{$_} ne $gst ) { - $conflict->{$group_readings->{$gst}} = $group_readings->{$contig{$_}}; + if( ref $contig->{$_} ) { + push( @{$contig->{$_}}, $gst ); + } elsif( $contig->{$_} ne $gst ) { + $conflict->{$group_readings->{$gst}} = $group_readings->{$contig->{$_}}; } # else it is an 'extant' node marked with our group already. } # None of the unreachable nodes should be in our group either. foreach ( $part->vertices ) { next if $reachable->{$_}; - if( $contig{$_} eq $gst ) { + if( $contig->{$_} eq $gst ) { $conflict->{$group_readings->{$gst}} = $group_readings->{$gst}; last; } @@ -245,7 +235,7 @@ sub analyze_variant_location { $reading->{'conflict'} = $conflict->{$group_readings->{$gst}} } else { # Save the relevant subgraph. - $subgraph{$gst} = { 'graph' => $part, + $subgraph->{$gst} = { 'graph' => $part, 'root' => $group_root, 'reachable' => $reachable }; } @@ -254,18 +244,18 @@ sub analyze_variant_location { # Now that we have gone through all the rows, check the hypothetical # readings for conflict if we haven't found one yet. - if( keys %subgraph && !keys %$conflict ) { + if( keys %$subgraph && !keys %$conflict ) { my @resolve; - foreach ( keys %contig ) { - next unless ref $contig{$_}; - if( scalar @{$contig{$_}} > 1 ) { + foreach ( keys %$contig ) { + next unless ref $contig->{$_}; + if( scalar @{$contig->{$_}} > 1 ) { push( @resolve, $_ ); } else { - $contig{$_} = scalar @{$contig{$_}} ? $contig{$_}->[0] : ''; + $contig->{$_} = scalar @{$contig->{$_}} ? $contig->{$_}->[0] : ''; } } # Do we still have a possible conflict? - my %still_contig; + my $still_contig = {}; foreach my $h ( @resolve ) { # For each of the hypothetical readings with more than one possibility, # try deleting it from each of its member subgraphs in turn, and see @@ -273,28 +263,28 @@ sub analyze_variant_location { # TODO This can still break in a corner case where group A can use # either vertex 1 or 2, and group B can use either vertex 2 or 1. # Revisit this if necessary; it could get brute-force nasty. - foreach my $gst ( @{$contig{$h}} ) { - my $gpart = $subgraph{$gst}->{'graph'}->copy; - my $reachable = $subgraph{$gst}->{'reachable'}; + foreach my $gst ( @{$contig->{$h}} ) { + my $gpart = $subgraph->{$gst}->{'graph'}->copy; + my $reachable = $subgraph->{$gst}->{'reachable'}; $gpart->delete_vertex( $h ); # Is everything else still reachable from the root? # TODO If $h was the root, see if we still have a single root. - my %still_reachable = ( $subgraph{$gst}->{'root'} => 1 ); + my %still_reachable = ( $subgraph->{$gst}->{'root'} => 1 ); map { $still_reachable{$_} = 1 } - $gpart->all_successors( $subgraph{$gst}->{'root'} ); + $gpart->all_successors( $subgraph->{$gst}->{'root'} ); foreach my $v ( keys %$reachable ) { next if $v eq $h; if( !$still_reachable{$v} - && ( $contig{$v} eq $gst - || ( exists $still_contig{$v} - && $still_contig{$v} eq $gst ) ) ) { + && ( $contig->{$v} eq $gst + || ( exists $still_contig->{$v} + && $still_contig->{$v} eq $gst ) ) ) { # We need $h. - if( exists $still_contig{$h} ) { + if( exists $still_contig->{$h} ) { # Conflict! $conflict->{$group_readings->{$gst}} = - $group_readings->{$still_contig{$h}}; + $group_readings->{$still_contig->{$h}}; } else { - $still_contig{$h} = $gst; + $still_contig->{$h} = $gst; } last; } # else we don't need $h in this group. @@ -306,9 +296,9 @@ sub analyze_variant_location { # $still_contig that are the "real" group memberships. Replace these # in $contig. unless ( keys %$conflict ) { - foreach my $v ( keys %contig ) { - next unless ref $contig{$v}; - $contig{$v} = $still_contig{$v}; + foreach my $v ( keys %$contig ) { + next unless ref $contig->{$v}; + $contig->{$v} = $still_contig->{$v}; } } } @@ -317,8 +307,8 @@ sub analyze_variant_location { foreach my $rdg ( @{$variant_row->{'readings'}} ) { $rdg->{'conflict'} = $conflict->{$rdg->{'text'}}; next if $rdg->{'conflict'}; - my @members = grep { $contig{$_} eq $rdg->{'group'} && !$missing{$_} } - keys %contig; + my @members = grep { $contig->{$_} eq $rdg->{'group'} && !$missing->{$_} } + keys %$contig; $rdg->{'group'} = wit_stringify( \@members ); } @@ -326,6 +316,26 @@ sub analyze_variant_location { return $variant_row; } +sub prune_subtree { + my( $tree, $root, $contighash ) = @_; + # First, delete hypothetical leaves / orphans until there are none left. + my @orphan_hypotheticals = grep { ref( $contighash->{$_} ) } + $tree->successorless_vertices; + while( @orphan_hypotheticals ) { + $tree->delete_vertices( @orphan_hypotheticals ); + @orphan_hypotheticals = grep { ref( $contighash->{$_} ) } + $tree->successorless_vertices; + } + # Then delete a hypothetical root with only one successor, moving the + # root to the child. + while( $tree->successors( $root ) == 1 && ref $contighash->{$root} ) { + my @nextroot = $tree->successors( $root ); + $tree->delete_vertex( $root ); + $root = $nextroot[0]; + } + # The tree has been modified in place, but we need to know the new root. + return $root; +} # Add the variant, subject to a.c. representation logic. # This assumes that we will see the 'main' version before the a.c. version. sub add_variant_wit {