use strict;
use warnings;
+use Benchmark;
+use Exporter 'import';
use Text::Tradition;
use Text::Tradition::Stemma;
-sub new {
- my( $class, $args ) = @_;
- my $self = {};
- # Our object needs to have a stemma graph and a variant table.
- my( $svg, $variants ) = run_analysis( $args->{'file'}, $args->{'stemmadot'} );
- $self->{'svg'} = $svg;
- $self->{'variants'} = $variants;
-
- bless( $self, $class );
- return $self;
-}
+use vars qw/ @EXPORT_OK /;
+@EXPORT_OK = qw/ run_analysis group_variants analyze_variant_location wit_stringify /;
+
+=head1 NAME
+
+Text::Tradition::Analysis - functions for stemma analysis of a tradition
+
+=head1 SYNOPSIS
+
+ use Text::Tradition;
+ use Text::Tradition::Analysis qw/ run_analysis analyze_variant_location /;
+ my $t = Text::Tradition->new(
+ 'name' => 'this is a text',
+ 'input' => 'TEI',
+ 'file' => '/path/to/tei_parallel_seg_file.xml' );
+ $t->add_stemma( 'dotfile' => $stemmafile );
+
+ my $variant_data = run_analysis( $tradition );
+ # Recalculate rank $n treating all orthographic variants as equivalent
+ my $reanalyze = analyze_variant_location( $tradition, $n, 0, 'orthographic' );
+
+=head1 DESCRIPTION
+
+Text::Tradition is a library for representation and analysis of collated
+texts, particularly medieval ones. The Collation is the central feature of
+a Tradition, where the text, its sequence of readings, and its relationships
+between readings are actually kept.
+
+=head1 SUBROUTINES
+
+=head2 run_analysis( $tradition, $stemma_id, @merge_relationship_types )
+
+Runs the analysis described in analyze_variant_location on every location
+in the collation of the given tradition, against the stemma specified in
+$stemma_id. If $stemma_id is not specified, it defaults to 0 (referencing
+the first stemma saved for the tradition.)
+
+The optional @merge_relationship_types contains a list of relationship types
+to treat as equivalent for the analysis.
+
+=begin testing
+
+use Text::Tradition;
+use Text::Tradition::Analysis qw/ run_analysis analyze_variant_location /;
+
+my $datafile = 't/data/florilegium_tei_ps.xml';
+my $tradition = Text::Tradition->new( 'input' => 'TEI',
+ 'name' => 'test0',
+ 'file' => $datafile );
+my $s = $tradition->add_stemma( 'dotfile' => 't/data/florilegium.dot' );
+is( ref( $s ), 'Text::Tradition::Stemma', "Added stemma to tradition" );
+
+my $data = run_analysis( $tradition );
+# TODO should be 21!
+is( $data->{'genealogical_count'}, 42, "Got right genealogical count" );
+is( $data->{'conflict_count'}, 17, "Got right conflict count" );
+is( $data->{'variant_count'}, 58, "Got right total variant number" );
+
+=end testing
+
+=cut
sub run_analysis {
- my( $file, $stemmadot ) = @_;
- # What we will return
- my $svg;
- my $variants = [];
+ my( $tradition, $stemma_id, @collapse ) = @_;
+ $stemma_id = 0 unless $stemma_id;
- # Read in the file and stemma
- my @lines;
- open( INFILE, "$file" ) or die "Could not read $file";
- binmode INFILE, ':utf8';
- @lines = <INFILE>;
- close INFILE;
+ # Run the variant analysis on every rank in the graph that doesn't
+ # have a common reading. Return the results.
+ my @variants; # holds results from analyze_variant_location
+ my $genealogical; # counter of 'genealogical' variants
+ my $conflicts; # counter of conflicting readings
- my $tradition = Text::Tradition->new(
- 'Self' => join( '', @lines ),
- 'linear' => 1,
- );
- my $stemma = Text::Tradition::Stemma->new(
- 'collation' => $tradition->collation,
- 'dot' => $stemmadot,
- );
- # We will return the stemma picture
- $svg = $stemma->as_svg;
- ### DIRTY HACK
- $svg =~ s/transform=\"scale\(1 1\)/transform=\"scale\(0.7 0.7\)/;
-
- # 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 = $tradition->collation->make_alignment_table( 'refs' );
+ # Find and mark 'common' ranks for exclusion.
+ my %common_rank;
+ foreach my $rdg ( $tradition->collation->common_readings ) {
+ $common_rank{$rdg->rank} = 1;
+ }
- # 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;
- my $html_data = [];
- my $total = 0; # Keep track of the total number of variant locations
+ foreach my $rank ( 1 .. $tradition->collation->end->rank-1 ) {
+ next if $common_rank{$rank};
+ my $variant_row = analyze_variant_location(
+ $tradition, $rank, $stemma_id, @collapse );
+ push( @variants, $variant_row );
+ $genealogical++ if $variant_row->{'genealogical'};
+ $conflicts += grep { $_->{'conflict'} } @{$variant_row->{'readings'}};
+ }
- # Strip the list of sigla and save it for correlation to the readings.
- my $col_wits = shift @$all_wits_table;
+ return {
+ 'variants' => \@variants,
+ 'variant_count' => scalar @variants, # TODO redundant
+ 'conflict_count' => $conflicts,
+ 'genealogical_count' => $genealogical,
+ };
+}
+
+=head2 group_variants( $tradition, $rank, $lacunose, @merge_relationship_types )
+
+Groups the variants at the given $rank of the collation, treating any
+relationships in @merge_relationship_types as equivalent. $lacunose should
+be a reference to an array, to which the sigla of lacunose witnesses at this
+rank will be appended.
+
+Returns two ordered lists $readings, $groups, where $readings->[$n] is attested
+by the witnesses listed in $groups->[$n].
+
+=cut
+
+# Return group_readings, groups, lacunose
+sub group_variants {
+ my( $tradition, $rank, $lacunose, $collapse ) = @_;
+ my $c = $tradition->collation;
+ # Get the alignment table readings
+ my %readings_at_rank;
+ my @gap_wits;
+ foreach my $tablewit ( @{$tradition->collation->alignment_table->{'alignment'}} ) {
+ my $rdg = $tablewit->{'tokens'}->[$rank-1];
+ if( $rdg && $rdg->{'t'}->is_lacuna ) {
+ push( @$lacunose, $tablewit->{'witness'} );
+ } elsif( $rdg ) {
+ $readings_at_rank{$rdg->{'t'}->text} = $rdg->{'t'};
+ } else {
+ push( @gap_wits, $tablewit->{'witness'} );
+ }
+ }
- # 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;
- foreach my $j ( 0 .. $#{$col_rdgs} ) {
- my $rdg = $col_rdgs->[$j];
- my $rdg_text = '(omitted)'; # Initialize in case of empty reading
- if( $rdg ) {
- $rdg_text = $rdg->is_lacuna ? undef : $rdg->text; # Don't count lacunae
- # Get the rank from any real reading; they should be identical.
- $rank = $rdg->rank unless $rank || $rdg->is_lacuna;
- }
- 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 );
+ # Group the readings, collapsing groups by relationship if needed
+ my %grouped_readings;
+ foreach my $rdg ( sort { $b->witnesses <=> $a->witnesses } values %readings_at_rank ) {
+ # Skip readings that have been collapsed into others.
+ next if exists $grouped_readings{$rdg->text} && !$grouped_readings{$rdg->text};
+ my @wits = $rdg->witnesses;
+ if( $collapse ) {
+ my $filter = sub { my $r = $_[0]; grep { $_ eq $r->type } @$collapse; };
+ foreach my $other ( $rdg->related_readings( $filter ) ) {
+ push( @wits, $other->witnesses );
+ $grouped_readings{$other->text} = 0;
}
}
-
- # 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;
-
- # Initialize the data structure for the row that we will return
- my $variant_row = {'id' => $rank, 'readings' => [] };
- # Keep track of our widest row
- $html_columns = scalar @$groups if scalar @$groups > $html_columns;
-
- # We can already look up witnesses for a reading; we also want to look
- # up readings for a given witness.
- my $group_readings = {};
- foreach my $x ( 0 .. $#$groups ) {
- $group_readings->{wit_stringify( $groups->[$x] )} = $readings->[$x];
- }
-
- # For all the groups with more than one member, collect the list of all
- # contiguous vertices needed to connect them.
- # TODO: deal with a.c. reading logic
- my $sc = analyze_variant_location( $group_readings, $groups, $stemma->apsp );
- $variant_row->{'genealogical'} = keys %$sc ? 1 : undef;
- foreach my $grp ( sort keys %$group_readings ) {
- my $rdg = $group_readings->{$grp};
- push( @{$variant_row->{'readings'}}, { 'text' => $rdg, 'group' => $grp } );
- }
-
- # Now run the same analysis given the calculated distance tree(s).
-# foreach my $tree ( 0 .. $#{$stemma->distance_trees} ) {
-# my $dc = analyze_variant_location( $group_readings, $groups,
-# $stemma->distance_apsps->[$tree] );
-# foreach my $rdg ( keys %$dc ) {
-# my $var = $dc->{$rdg};
-# }
-# }
-
- # Record that we used this variant in an analysis
- push( @$variants, $variant_row );
+ $grouped_readings{$rdg->text} = \@wits;
}
+ $grouped_readings{'(omitted)'} = \@gap_wits if @gap_wits;
+ # Get rid of our collapsed readings
+ map { delete $grouped_readings{$_} unless $grouped_readings{$_} }
+ keys %grouped_readings
+ if $collapse;
- # Go through our variant rows and add the number of empty columns we need.
- foreach my $row ( @$variants ) {
- my $empty = $html_columns - scalar @{$row->{'readings'}};
- $row->{'empty'} = $empty;
+ # Return the readings and groups, sorted by size
+ my( @readings, @groups );
+ foreach my $r ( sort { @{$grouped_readings{$b}} <=> @{$grouped_readings{$a}} }
+ keys %grouped_readings ) {
+ push( @readings, $r );
+ push( @groups, $grouped_readings{$r} );
}
-
- return( $svg, $variants );
+ return( \@readings, \@groups );
}
+=head2 analyze_variant_location( $tradition, $rank, $stemma_id, @merge_relationship_types )
+
+Runs an analysis of the given tradition, at the location given in $rank,
+against the graph of the stemma specified in $stemma_id. The argument
+@merge_relationship_types is an optional list of relationship types for
+which readings so related should be treated as equivalent.
+
+Returns a data structure as follows:
+
+ { 'id' => $rank,
+ 'genealogical' => boolean,
+ 'readings => [ { text => $reading_text,
+ group => [ witnesses ],
+ conflict => [ conflicting ],
+ missing => [ excluded ] }, ... ]
+ }
+where 'conflicting' is the list of witnesses whose readings conflict with
+this group, and 'excluded' is the list of witnesses either not present in
+the stemma or lacunose at this location.
+
+=cut
+
sub analyze_variant_location {
- my( $group_readings, $groups, $apsp ) = @_;
- my %contig;
+ my( $tradition, $rank, $sid, @collapse ) = @_;
+ $DB::single = 1 if @collapse;
+ # Get the readings in this tradition at this rank
+ my @rank_rdgs = grep { $_->rank == $rank } $tradition->collation->readings;
+ # Get the applicable stemma
+ my $undirected; # TODO Allow undirected distance tree analysis too
+ my $stemma = $tradition->stemma( $sid );
+ my $graph = $stemma->graph;
+ # Figure out which witnesses we are working with
+ my @lacunose = _set( 'symmdiff', [ $stemma->witnesses ],
+ [ map { $_->sigil } $tradition->witnesses ] );
+
+ # Now group the readings
+ my( $readings, $groups ) =
+ group_variants( $tradition, $rank, \@lacunose, \@collapse );
+ my $group_readings = {};
+ # Lookup table group string -> readings
+ foreach my $x ( 0 .. $#$groups ) {
+ $group_readings->{wit_stringify( $groups->[$x] )} = $readings->[$x];
+ }
+
+ # Now do the work.
+ my $contig = {};
+ my $subgraph = {};
+ my $is_conflicted;
my $conflict = {};
- foreach my $g ( sort { scalar @$b <=> scalar @$a } @$groups ) {
- my @members = @$g;
+ my $variant_row = { 'id' => $rank, 'readings' => [] };
+ # Mark each ms as in its own group, first.
+ foreach my $g ( @$groups ) {
my $gst = wit_stringify( $g );
- map { $contig{$_} = $gst } @members; # The witnesses need themselves to be
- # in their collection.
- next unless @members > 1;
- my $curr = pop @members;
- foreach my $m ( @members ) {
- foreach my $v ( $apsp->path_vertices( $curr, $m ) ) {
- $contig{$v} = $gst unless exists $contig{$v};
- next if $contig{$v} eq $gst;
- # print STDERR "Conflict at $v between group $gst and group "
- # . $contig{$v} . "\n";
- # Record what is conflicting.
- $conflict->{$group_readings->{$gst}} = $group_readings->{$contig{$v}};
+ 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;
+ foreach my $g ( sort { scalar @$b <=> scalar @$a } @$groups ) {
+ my $gst = wit_stringify( $g ); # This is the group name
+ my $reachable = { $g->[0] => 1 };
+ # Copy the graph, and delete all non-members from the new graph.
+ my $part = $graph->copy;
+ my $group_root;
+ $part->delete_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.
+ # Find all vertices reachable from the first (arbitrary) group
+ # member. If we are genealogical this should include them all.
+ map { $reachable->{$_} = 1 } $part->all_reachable( $g->[0] );
+ # TODO This is a terrible way to do distance trees, since all
+ # non-leaf nodes are included in every graph part now. We may
+ # have to go back to SPDP.
+ } else {
+ if( @$g > 1 ) {
+ # 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 }
+ $part->source_vertices;
+ # Assuming that @$g > 1, find the first root node that has at
+ # least one successor belonging to our group. If this reading
+ # is genealogical, there should be only one, but we will check
+ # that implicitly later.
+ my $nodes_in_subtree = 0;
+ foreach my $root ( @roots ) {
+ # 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 %$tmp_reach;
+ if( keys %$tmp_reach > $nodes_in_subtree ) {
+ $nodes_in_subtree = keys %$tmp_reach;
+ $reachable = $tmp_reach;
+ $group_root = $root;
+ }
+ }
+ } # else it is a single-node group, nothing to calculate.
+ }
+
+ # None of the 'reachable' nodes should be marked as being in another
+ # group. Paint the 'hypotheticals' with our group while we are at it,
+ # unless there is a conflict present.
+ foreach ( keys %$reachable ) {
+ if( ref $contig->{$_} ) {
+ push( @{$contig->{$_}}, $gst );
+ } elsif( $contig->{$_} ne $gst ) {
+ $conflict->{$group_readings->{$gst}} = $group_readings->{$contig->{$_}};
+ } # 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 ) {
+ $conflict->{$group_readings->{$gst}} = $group_readings->{$gst};
+ last;
}
}
+
+ # Now, if we have a conflict, we can write the reading in full. If not,
+ # we have to save the subgraph so that we can resolve possible conflicts
+ # on hypothetical nodes.
+ $is_conflicted = 1 if exists $conflict->{$group_readings->{$gst}};
+
+ # Write the reading.
+ my $reading = { 'text' => $group_readings->{$gst},
+ 'missing' => wit_stringify( \@lacunose ),
+ 'group' => $gst }; # This will change if we find no conflict
+ if( $is_conflicted ) {
+ $reading->{'conflict'} = $conflict->{$group_readings->{$gst}}
+ } else {
+ # Save the relevant subgraph.
+ $subgraph->{$gst} = { 'graph' => $part,
+ 'root' => $group_root,
+ 'reachable' => $reachable };
+ }
+ push( @{$variant_row->{'readings'}}, $reading );
}
- return $conflict;
+
+ # Now that we have gone through all the rows, check the hypothetical
+ # readings for conflict if we haven't found one yet.
+ if( keys %$subgraph && !keys %$conflict ) {
+ my @resolve;
+ foreach ( keys %$contig ) {
+ next unless ref $contig->{$_};
+ if( scalar @{$contig->{$_}} > 1 ) {
+ push( @resolve, $_ );
+ } else {
+ $contig->{$_} = scalar @{$contig->{$_}} ? $contig->{$_}->[0] : '';
+ }
+ }
+ # Do we still have a possible conflict?
+ 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
+ # if that breaks the contiguous grouping.
+ # 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'};
+ $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 );
+ map { $still_reachable{$_} = 1 }
+ $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 ) ) ) {
+ # We need $h.
+ if( exists $still_contig->{$h} ) {
+ # Conflict!
+ $conflict->{$group_readings->{$gst}} =
+ $group_readings->{$still_contig->{$h}};
+ } else {
+ $still_contig->{$h} = $gst;
+ }
+ last;
+ } # else we don't need $h in this group.
+ }
+ }
+ }
+
+ # Now, assuming no conflict, we have some hypothetical vertices in
+ # $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};
+ }
+ }
+ }
+
+ # Now write the group and conflict information into the respective rows.
+ my %missing;
+ map { $missing{$_} = 1 } @lacunose; # quick lookup table
+ 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;
+ $rdg->{'group'} = wit_stringify( \@members );
+ }
+
+ $variant_row->{'genealogical'} = !( keys %$conflict );
+ 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 $root;
+}
# Add the variant, subject to a.c. representation logic.
# This assumes that we will see the 'main' version before the a.c. version.
sub add_variant_wit {
push( @$arr, $wit ) unless $skip;
}
-# Return an answer if the variant is useful, i.e. if there are at least 2 variants
-# with at least 2 witnesses each.
-sub useful_variant {
- my( $readings ) = @_;
- my $total = keys %$readings;
- foreach my $var ( keys %$readings ) {
- $total-- if @{$readings->{$var}} == 1;
- }
- return( undef, undef ) if $total <= 1;
- my( $groups, $text );
- foreach my $var ( keys %$readings ) {
- push( @$groups, $readings->{$var} );
- push( @$text, $var );
- }
- return( $groups, $text );
-}
+=head2 wit_stringify( $groups )
+
+Takes an array of witness groupings and produces a string like
+['A','B'] / ['C','D','E'] / ['F']
-# Take an array of witness groupings and produce a string like
-# ['A','B'] / ['C','D','E'] / ['F']
+=cut
sub wit_stringify {
my $groups = shift;
}
return join( ' / ', @gst );
}
-
-1;
\ No newline at end of file
+
+sub _set {
+ my( $op, $lista, $listb ) = @_;
+ my %union;
+ my %scalars;
+ map { $union{$_} = 1; $scalars{$_} = $_ } @$lista;
+ map { $union{$_} += 1; $scalars{$_} = $_ } @$listb;
+ my @set;
+ if( $op eq 'intersection' ) {
+ @set = grep { $union{$_} == 2 } keys %union;
+ } elsif( $op eq 'symmdiff' ) {
+ @set = grep { $union{$_} == 1 } keys %union;
+ } elsif( $op eq 'union' ) {
+ @set = keys %union;
+ }
+ return map { $scalars{$_} } @set;
+}
+
+1;
+
+=head1 LICENSE
+
+This package is free software and is provided "as is" without express
+or implied warranty. You can redistribute it and/or modify it under
+the same terms as Perl itself.
+
+=head1 AUTHOR
+
+Tara L Andrews E<lt>aurum@cpan.orgE<gt>