use strict;
use warnings;
use Benchmark;
+use Encode qw/ encode_utf8 /;
use Exporter 'import';
+use JSON qw/ encode_json decode_json /;
+use LWP::UserAgent;
use Text::Tradition;
use Text::Tradition::Stemma;
=head1 SUBROUTINES
-=head2 run_analysis( $tradition, $stemma_id, @merge_relationship_types )
+=head2 run_analysis( $tradition, %opts )
-Runs the analysis described in analyze_variant_location on every location
-in the collation of the given tradition, against the stemma specified in
-$stemma_id. If $stemma_id is not specified, it defaults to 0 (referencing
-the first stemma saved for the tradition.)
+Runs the analysis described in analyze_variant_location on every location in the
+collation of the given tradition, with the given options. These include:
-The optional @merge_relationship_types contains a list of relationship types
-to treat as equivalent for the analysis.
+=over 4
+
+=item * stemma_id - Specify which of the tradition's stemmata to use. Default
+is 0 (i.e. the first).
+
+=item * ranks - Specify a list of location ranks to analyze; exclude the rest.
+
+=item * merge_types - Specify a list of relationship types, where related readings
+should be treated as identical for the purposes of analysis.
+
+=back
=begin testing
is( ref( $s ), 'Text::Tradition::Stemma', "Added stemma to tradition" );
my %expected_genealogical = (
- 1 => '',
+ 1 => 0,
2 => 1,
- 3 => '',
- 5 => '',
- 7 => '',
- 8 => '',
- 10 => '',
+ 3 => 0,
+ 5 => 0,
+ 7 => 0,
+ 8 => 0,
+ 10 => 0,
13 => 1,
- 33 => '',
- 34 => '',
- 37 => '',
- 60 => '',
+ 33 => 0,
+ 34 => 0,
+ 37 => 0,
+ 60 => 0,
81 => 1,
- 84 => '',
- 87 => '',
- 101 => '',
- 102 => '',
+ 84 => 0,
+ 87 => 0,
+ 101 => 0,
+ 102 => 0,
122 => 1,
- 157 => '',
+ 157 => 0,
166 => 1,
169 => 1,
- 200 => 1,
+ 200 => 0,
216 => 1,
217 => 1,
219 => 1,
my $data = run_analysis( $tradition );
foreach my $row ( @{$data->{'variants'}} ) {
+ # Account for rows that used to be "not useful"
+ unless( exists $expected_genealogical{$row->{'id'}} ) {
+ $expected_genealogical{$row->{'id'}} = 1;
+ }
is( $row->{'genealogical'}, $expected_genealogical{$row->{'id'}},
"Got correct genealogical flag for row " . $row->{'id'} );
}
-is( $data->{'conflict_count'}, 16, "Got right conflict count" );
-is( $data->{'variant_count'}, 28, "Got right total variant number" );
+is( $data->{'conflict_count'}, 34, "Got right conflict count" );
+is( $data->{'variant_count'}, 58, "Got right total variant number" );
=end testing
=cut
sub run_analysis {
- my( $tradition, $stemma_id, @collapse ) = @_;
+ my( $tradition, %opts ) = @_;
my $c = $tradition->collation;
- $stemma_id = 0 unless $stemma_id;
-
- # 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 $stemma_id = $opts{'stemma_id'} || 0;
+ my @ranks = @{$opts{'ranks'}} if ref( $opts{'ranks'} ) eq 'ARRAY';
+ my @collapse = @{$opts{'merge_types'}} if ref( $opts{'merge_types'} ) eq 'ARRAY';
+
+ # Get the stemma
+ my $stemma = $tradition->stemma( $stemma_id );
+ # Figure out which witnesses we are working with
+ my @lacunose = $stemma->hypotheticals;
+ my @tradition_wits = map { $_->sigil } $tradition->witnesses;
+ map { push( @tradition_wits, $_->sigil."_ac" ) if $_->is_layered }
+ $tradition->witnesses;
+ push( @lacunose, _symmdiff( [ $stemma->witnesses ], \@tradition_wits ) );
+
+ # Find and mark 'common' ranks for exclusion, unless they were
+ # explicitly specified.
+ unless( @ranks ) {
+ my %common_rank;
+ foreach my $rdg ( $c->common_readings ) {
+ $common_rank{$rdg->rank} = 1;
+ }
+ @ranks = grep { !$common_rank{$_} } ( 1 .. $c->end->rank-1 );
+ }
- # Find and mark 'common' ranks for exclusion.
- my %common_rank;
- foreach my $rdg ( $tradition->collation->common_readings ) {
- $common_rank{$rdg->rank} = 1;
+ # Group the variants to send to the solver
+ my @groups;
+ my %lacunae;
+ foreach my $rank ( @ranks ) {
+ my $missing = [ @lacunose ];
+ push( @groups, group_variants( $tradition, $rank, $missing, \@collapse ) );
+ $lacunae{$rank} = $missing;
}
- 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 );
- next unless $variant_row;
- # Add the reading text to the readings in variant_row
- foreach my $rh ( @{$variant_row->{'readings'}} ) {
- if( $c->reading( $rh->{'readingid'} ) ) {
- $rh->{'text'} = $c->reading( $rh->{'readingid'} )->text;
- } else {
- $rh->{'text'} = $rh->{'readingid'};
- }
+ # Parse the answer
+ my $answer = solve_variants( $stemma->editable( ' ' ), @groups );
+
+ # Do further analysis on the answer
+ my $conflict_count = 0;
+ foreach my $idx ( 0 .. $#ranks ) {
+ my $location = $answer->{'variants'}->[$idx];
+ # Add the rank back in
+ $location->{'id'} = $ranks[$idx];
+ # Add the lacunae back in
+ $location->{'missing'} = $lacunae{$ranks[$idx]};
+ # Run the extra analysis we need.
+ analyze_location( $tradition, $stemma->graph, $location );
+ # Add the reading text back in
+ foreach my $rdghash ( @{$location->{'readings'}} ) {
+ $conflict_count++
+ if exists $rdghash->{'conflict'} && $rdghash->{'conflict'};
+ my $rdg = $c->reading( $rdghash->{'readingid'} );
+ $rdghash->{'text'} = $rdg ? $rdg->text : $rdghash->{'readingid'};
}
- push( @variants, $variant_row );
- $genealogical++ if $variant_row->{'genealogical'};
- $conflicts += grep { $_->{'conflict'} } @{$variant_row->{'readings'}};
}
+ $answer->{'conflict_count'} = $conflict_count;
-
- return {
- 'variants' => \@variants,
- 'variant_count' => scalar @variants, # TODO redundant
- 'conflict_count' => $conflicts,
- 'genealogical_count' => $genealogical,
- };
+ return $answer;
}
=head2 group_variants( $tradition, $rank, $lacunose, @merge_relationship_types )
sub group_variants {
my( $tradition, $rank, $lacunose, $collapse ) = @_;
my $c = $tradition->collation;
+ my $aclabel = $c->ac_label;
# 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];
+ my $wit = $tablewit->{'witness'};
+ $wit =~ s/^(.*)\Q$aclabel\E$/${1}_ac/;
if( $rdg && $rdg->{'t'}->is_lacuna ) {
- _add_to_witlist( $tablewit->{'witness'}, $lacunose,
- $tradition->collation->ac_label );
+ _add_to_witlist( $wit, $lacunose, '_ac' );
} elsif( $rdg ) {
$readings_at_rank{$rdg->{'t'}->text} = $rdg->{'t'};
} else {
- _add_to_witlist( $tablewit->{'witness'}, \@gap_wits,
- $tradition->collation->ac_label );
+ _add_to_witlist( $wit, \@gap_wits, '_ac' );
}
}
# Skip readings that have been collapsed into others.
next if exists $grouped_readings{$rdg->id} && !$grouped_readings{$rdg->id};
my @wits = $rdg->witnesses;
+ map { s/\Q$aclabel\E$/_ac/ } @wits;
if( $collapse ) {
my $filter = sub { my $r = $_[0]; grep { $_ eq $r->type } @$collapse; };
foreach my $other ( $rdg->related_readings( $filter ) ) {
- push( @wits, $other->witnesses );
+ my @otherwits = $other->witnesses;
+ map { s/\Q$aclabel\E$/_ac/ } @otherwits;
+ push( @wits, @otherwits );
$grouped_readings{$other->id} = 0;
}
}
return \%grouped_readings;
}
-=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:
+=head2 solve_variants( $graph, @groups )
- { 'id' => $rank,
- 'genealogical' => boolean,
- 'readings => [ { readingid => $reading_id,
- 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.
+Sends the set of groups to the external graph solver service and returns
+a cleaned-up answer, adding the rank IDs back where they belong.
+The JSON has the form
+ { "graph": [ stemmagraph DOT string without newlines ],
+ "groupings": [ array of arrays of groups, one per rank ] }
+
+The answer has the form
+ { "variants" => [ array of variant location structures ],
+ "variant_count" => total,
+ "conflict_count" => number of conflicts detected,
+ "genealogical_count" => number of solutions found }
+
=cut
-sub analyze_variant_location {
- my( $tradition, $rank, $sid, @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 = $stemma->hypotheticals;
- push( @lacunose, _symmdiff( [ $stemma->witnesses ],
- [ map { $_->sigil } $tradition->witnesses ] ) );
-
- # Now group the readings
- my( $readings, $groups ) = _useful_variant(
- group_variants( $tradition, $rank, \@lacunose, \@collapse ),
- $graph, $tradition->collation->ac_label );
- return unless scalar @$readings;
- my $group_readings = {};
- # Lookup table group string -> readings
- foreach my $x ( 0 .. $#$groups ) {
- $group_readings->{wit_stringify( $groups->[$x] )} = $readings->[$x];
- }
+sub solve_variants {
+ my( $graph, @groups ) = @_;
- # Now do the work.
- my $contig = {};
- my $subgraph = {};
- my $is_conflicted;
- my $conflict = {};
- my %reading_roots;
- my $variant_row = { 'id' => $rank, 'readings' => [] };
- # Mark each ms as in its own group, first.
- $DB::single = 1 if $rank == 81;
- foreach my $g ( @$groups ) {
- my $gst = wit_stringify( $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;
- foreach my $g ( sort { scalar @$b <=> scalar @$a } @$groups ) {
- my $gst = wit_stringify( $g ); # This is the group name
- # Copy the graph, and delete all non-members from the new graph.
- my $part = $graph->copy;
- my @group_roots;
- $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.
- my $reachable = {};
- 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 ) {
- # We have to take directionality into account.
- # How many root nodes do we have?
- my @roots = grep { ref( $contig->{$_} ) || $contig->{$_} eq $gst }
- $part->predecessorless_vertices;
- # Assuming that @$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.
- foreach my $root ( @roots ) {
- # Prune the tree to get rid of extraneous hypotheticals.
- $root = _prune_subtree( $part, $root, $contig );
- next unless $root;
- # Save this root for our group.
- push( @group_roots, $root );
- # Get all the successor nodes of our root.
- }
- } else {
- # Dispense with the trivial case of one reading.
- my $wit = pop @$g;
- @group_roots = ( $wit );
- foreach my $v ( $part->vertices ) {
- $part->delete_vertex( $v ) unless $v eq $wit;
- }
- }
- }
-
- map { $reading_roots{$_} = 1 } @group_roots;
- if( @group_roots > 1 ) {
- $conflict->{$group_readings->{$gst}} = 1;
- $is_conflicted = 1;
- }
- # Paint the 'hypotheticals' with our group.
- foreach my $wit ( $part->vertices ) {
- if( ref( $contig->{$wit} ) ) {
- push( @{$contig->{$wit}}, $gst );
- } elsif( $contig->{$wit} ne $gst ) {
- warn "How did we get here?";
- }
+ # Make the json with stemma + groups
+ my $jsonstruct = { 'graph' => $graph, 'groupings' => [] };
+ foreach my $ghash ( @groups ) {
+ my @grouping;
+ foreach my $k ( sort keys %$ghash ) {
+ push( @grouping, $ghash->{$k} );
}
-
-
- # Start to write the reading, and save the group subgraph.
- my $reading = { 'readingid' => $group_readings->{$gst},
- 'missing' => wit_stringify( \@lacunose ),
- 'group' => $gst }; # This will change if we find no conflict
- # Save the relevant subgraph.
- $subgraph->{$gst} = $part;
- push( @{$variant_row->{'readings'}}, $reading );
- }
-
- # For each of our hypothetical readings, flatten its 'contig' array if
- # the array contains zero or one group. If we have any unflattened arrays,
- # we may need to run the resolution process. If the reading is already known
- # to have a conflict, flatten the 'contig' array to nothing; we won't resolve
- # it.
- my @resolve;
- foreach my $wit ( keys %$contig ) {
- next unless ref( $contig->{$wit} );
- if( @{$contig->{$wit}} > 1 ) {
- if( $is_conflicted ) {
- $contig->{$wit} = ''; # We aren't going to decide.
- } else {
- push( @resolve, $wit );
- }
- } else {
- my $gst = pop @{$contig->{$wit}};
- $contig->{$wit} = $gst || '';
+ push( @{$jsonstruct->{'groupings'}}, \@grouping );
+ }
+ my $json = encode_json( $jsonstruct );
+
+ # Send it off and get the result
+ my $solver_url = 'http://byzantini.st/cgi-bin/graphcalc.cgi';
+ my $ua = LWP::UserAgent->new();
+ my $resp = $ua->post( $solver_url, 'Content-Type' => 'application/json',
+ 'Content' => $json );
+
+ my $answer;
+ if( $resp->is_success ) {
+ $answer = decode_json( $resp->content );
+ } else {
+ # Fall back to the old method.
+ warn "IDP solver returned " . $resp->status_line . " / " . $resp->content
+ . "; falling back to perl method";
+ $answer = perl_solver( $graph, @groups );
+ }
+
+ # Fold the result back into what we know about the groups.
+ my $variants = [];
+ my $genealogical = 0;
+ foreach my $idx ( 0 .. $#groups ) {
+ my( $calc_groups, $result ) = @{$answer->[$idx]};
+ $genealogical++ if $result;
+ my $input_group = $groups[$idx];
+ foreach my $k ( sort keys %$input_group ) {
+ my $cg = shift @$calc_groups;
+ $input_group->{$k} = $cg;
}
+ my $vstruct = {
+ 'genealogical' => $result,
+ 'readings' => [],
+ };
+ foreach my $k ( keys %$input_group ) {
+ push( @{$vstruct->{'readings'}},
+ { 'readingid' => $k, 'group' => $input_group->{$k}} );
+ }
+ push( @$variants, $vstruct );
}
- if( @resolve ) {
- 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}->copy();
- # If we have come this far, there is only one root and everything
- # is reachable from it.
- my( $root ) = $gpart->predecessorless_vertices;
- my $reachable = {};
- map { $reachable->{$_} = 1 } $gpart->vertices;
-
- # Try deleting the hypothetical node.
- $gpart->delete_vertex( $h );
- if( $h eq $root ) {
- # See if we still have a single root.
- my @roots = $gpart->predecessorless_vertices;
- warn "This shouldn't have happened" unless @roots;
- if( @roots > 1 ) {
- # $h is needed by this group.
- if( exists( $still_contig->{$h} ) ) {
- # Conflict!
- $conflict->{$group_readings->{$gst}} = 1;
- $still_contig->{$h} = '';
- } else {
- $still_contig->{$h} = $gst;
- }
- }
- } else {
- # $h is somewhere in the middle. See if everything
- # else can still be reached from the root.
- my %still_reachable = ( $root => 1 );
- map { $still_reachable{$_} = 1 }
- $gpart->all_successors( $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}} = 1;
- $still_contig->{$h} = '';
- } else {
- $still_contig->{$h} = $gst;
- }
- last;
- } # else we don't need $h in this group.
- } # end foreach $v
- } # endif $h eq $root
- } # end foreach $gst
- } # end foreach $h
+ return { 'variants' => $variants,
+ 'variant_count' => scalar @$variants,
+ 'genealogical_count' => $genealogical };
+}
+
+=head2 analyze_location ( $tradition, $graph, $location_hash )
+
+Given the tradition, its stemma graph, and the solution from the graph solver,
+work out the rest of the information we want. For each reading we need missing,
+conflict, reading_parents, independent_occurrence, followed, not_followed, and follow_unknown. Alters the location_hash in place.
+
+=cut
+
+sub analyze_location {
+ my ( $tradition, $graph, $variant_row ) = @_;
+
+ # Make a hash of all known node memberships, and make the subgraphs.
+ my $contig = {};
+ my $reading_roots = {};
+ my $subgraph = {};
+ foreach my $rdghash ( @{$variant_row->{'readings'}} ) {
+ my $rid = $rdghash->{'readingid'};
+ map { $contig->{$_} = $rid } @{$rdghash->{'group'}};
- # Now we have some hypothetical vertices in $still_contig that are the
- # "real" group memberships. Replace these in $contig.
- foreach my $v ( keys %$contig ) {
- next unless ref $contig->{$v};
- $contig->{$v} = $still_contig->{$v};
- }
- } # end if @resolve
-
- # Now that we have all the node group memberships, calculate followed/
+ # Make the subgraph.
+ my $part = $graph->copy;
+ my %these_vertices;
+ map { $these_vertices{$_} = 1 } @{$rdghash->{'group'}};
+ $part->delete_vertices( grep { !$these_vertices{$_} } $part->vertices );
+ $subgraph->{$rid} = $part;
+ # Get the reading roots.
+ map { $reading_roots->{$_} = $rid } $part->predecessorless_vertices;
+ }
+
+ # Now that we have all the node group memberships, calculate followed/
# non-followed/unknown values for each reading. Also figure out the
# reading's evident parent(s).
foreach my $rdghash ( @{$variant_row->{'readings'}} ) {
- my $gst = $rdghash->{'group'};
- my $part = $subgraph->{$gst};
+ # Group string key - TODO do we need this?
+ my $gst = wit_stringify( $rdghash->{'group'} );
+ my $rid = $rdghash->{'readingid'};
+ # Get the subgraph
+ my $part = $subgraph->{$rid};
+
+ # Start figuring things out.
my @roots = $part->predecessorless_vertices;
$rdghash->{'independent_occurrence'} = scalar @roots;
$rdghash->{'followed'} = scalar( $part->vertices ) - scalar( @roots );
while( @check ) {
my @next;
foreach my $wparent( @check ) {
- my $pgroup = $contig->{$wparent};
- if( $pgroup ) {
- $rdgparents{$group_readings->{$pgroup}} = 1;
+ my $preading = $contig->{$wparent};
+ if( $preading ) {
+ $rdgparents{$preading} = 1;
} else {
push( @next, $graph->predecessors( $wparent ) );
}
foreach my $wit ( $part->vertices ) {
foreach my $wchild ( $graph->successors( $wit ) ) {
next if $part->has_vertex( $wchild );
- if( $reading_roots{$wchild} && $contig->{$wchild} ) {
+ if( $reading_roots->{$wchild} && $contig->{$wchild} ) {
# It definitely changed here.
$nofollow{$wchild} = 1;
} elsif( !($contig->{$wchild}) ) {
}
$rdghash->{'not_followed'} = keys %nofollow;
$rdghash->{'follow_unknown'} = keys %unknownfollow;
+
+ # Now say whether this reading represents a conflict.
+ unless( $variant_row->{'genealogical'} ) {
+ $rdghash->{'conflict'} = @roots != 1;
+ }
}
-
- # Now write the group and conflict information into the respective rows.
- foreach my $rdghash ( @{$variant_row->{'readings'}} ) {
- $rdghash->{'conflict'} = $conflict->{$rdghash->{'readingid'}};
- my @members = grep { $contig->{$_} eq $rdghash->{'group'} } keys %$contig;
- $rdghash->{'group'} = wit_stringify( \@members );
- }
-
- $variant_row->{'genealogical'} = !( keys %$conflict );
- return $variant_row;
}
+
+=head2 perl_solver( $tradition, $rank, $stemma_id, @merge_relationship_types )
+
+** NOTE ** This method should hopefully not be called - it is not guaranteed
+to be correct. Serves as a backup for the real solver.
+
+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 nested array data structure as follows:
+
+ [ [ group_list, is_genealogical ], [ group_list, is_genealogical ] ... ]
+
+where the group list is the array of arrays passed in for each element of @groups,
+possibly with the addition of hypothetical readings.
+
+
+=cut
+
+sub perl_solver {
+ my( $graph, @groups ) = @_;
+
+ warn "Not implemented yet";
+ return [];
+}
+
+ # Now do the work.
+# my $contig = {};
+# my $subgraph = {};
+# my $is_conflicted;
+# my $conflict = {};
+# my %reading_roots;
+# 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 { $contig->{$_} = $gst } @$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
+# # Copy the graph, and delete all non-members from the new graph.
+# my $part = $graph->copy;
+# my @group_roots;
+# $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.
+# my $reachable = {};
+# 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 ) {
+# # We have to take directionality into account.
+# # How many root nodes do we have?
+# my @roots = grep { ref( $contig->{$_} ) || $contig->{$_} eq $gst }
+# $part->predecessorless_vertices;
+# # Assuming that @$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.
+# foreach my $root ( @roots ) {
+# # Prune the tree to get rid of extraneous hypotheticals.
+# $root = _prune_subtree( $part, $root, $contig );
+# next unless $root;
+# # Save this root for our group.
+# push( @group_roots, $root );
+# # Get all the successor nodes of our root.
+# }
+# } else {
+# # Dispense with the trivial case of one reading.
+# my $wit = pop @$g;
+# @group_roots = ( $wit );
+# foreach my $v ( $part->vertices ) {
+# $part->delete_vertex( $v ) unless $v eq $wit;
+# }
+# }
+# }
+#
+# map { $reading_roots{$_} = 1 } @group_roots;
+# if( @group_roots > 1 ) {
+# $conflict->{$group_readings->{$gst}} = 1;
+# $is_conflicted = 1;
+# }
+# # Paint the 'hypotheticals' with our group.
+# foreach my $wit ( $part->vertices ) {
+# if( ref( $contig->{$wit} ) ) {
+# push( @{$contig->{$wit}}, $gst );
+# } elsif( $contig->{$wit} ne $gst ) {
+# warn "How did we get here?";
+# }
+# }
+#
+#
+# # Start to write the reading, and save the group subgraph.
+# my $reading = { 'readingid' => $group_readings->{$gst},
+# 'missing' => wit_stringify( \@lacunose ),
+# 'group' => $gst }; # This will change if we find no conflict
+# # Save the relevant subgraph.
+# $subgraph->{$gst} = $part;
+# push( @{$variant_row->{'readings'}}, $reading );
+# }
+#
+# # For each of our hypothetical readings, flatten its 'contig' array if
+# # the array contains zero or one group. If we have any unflattened arrays,
+# # we may need to run the resolution process. If the reading is already known
+# # to have a conflict, flatten the 'contig' array to nothing; we won't resolve
+# # it.
+# my @resolve;
+# foreach my $wit ( keys %$contig ) {
+# next unless ref( $contig->{$wit} );
+# if( @{$contig->{$wit}} > 1 ) {
+# if( $is_conflicted ) {
+# $contig->{$wit} = ''; # We aren't going to decide.
+# } else {
+# push( @resolve, $wit );
+# }
+# } else {
+# my $gst = pop @{$contig->{$wit}};
+# $contig->{$wit} = $gst || '';
+# }
+# }
+#
+# if( @resolve ) {
+# 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}->copy();
+# # If we have come this far, there is only one root and everything
+# # is reachable from it.
+# my( $root ) = $gpart->predecessorless_vertices;
+# my $reachable = {};
+# map { $reachable->{$_} = 1 } $gpart->vertices;
+#
+# # Try deleting the hypothetical node.
+# $gpart->delete_vertex( $h );
+# if( $h eq $root ) {
+# # See if we still have a single root.
+# my @roots = $gpart->predecessorless_vertices;
+# warn "This shouldn't have happened" unless @roots;
+# if( @roots > 1 ) {
+# # $h is needed by this group.
+# if( exists( $still_contig->{$h} ) ) {
+# # Conflict!
+# $conflict->{$group_readings->{$gst}} = 1;
+# $still_contig->{$h} = '';
+# } else {
+# $still_contig->{$h} = $gst;
+# }
+# }
+# } else {
+# # $h is somewhere in the middle. See if everything
+# # else can still be reached from the root.
+# my %still_reachable = ( $root => 1 );
+# map { $still_reachable{$_} = 1 }
+# $gpart->all_successors( $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}} = 1;
+# $still_contig->{$h} = '';
+# } else {
+# $still_contig->{$h} = $gst;
+# }
+# last;
+# } # else we don't need $h in this group.
+# } # end foreach $v
+# } # endif $h eq $root
+# } # end foreach $gst
+# } # end foreach $h
+#
+# # Now we have some hypothetical vertices in $still_contig that are the
+# # "real" group memberships. Replace these in $contig.
+# foreach my $v ( keys %$contig ) {
+# next unless ref $contig->{$v};
+# $contig->{$v} = $still_contig->{$v};
+# }
+# } # end if @resolve
+#
+#
+# $variant_row->{'genealogical'} = !( keys %$conflict );
+# return $variant_row;
+# }
+
sub _prune_subtree {
my( $tree, $root, $contighash ) = @_;
# First, delete hypothetical leaves / orphans until there are none left.
projects / scpubgit/stemmatology.git / blobdiff