use strict;
use warnings;
-use Benchmark;
-use Encode qw/ encode_utf8 /;
+use Algorithm::Diff; # for word similarity measure
+use Encode qw/ decode_utf8 encode_utf8 /;
use Exporter 'import';
use Graph;
-use JSON qw/ encode_json decode_json /;
+use JSON qw/ to_json decode_json /;
use LWP::UserAgent;
-use Text::Tradition;
+use Set::Scalar;
+use Text::Tradition::Analysis::Result;
+use Text::Tradition::Directory;
use Text::Tradition::Stemma;
+use TryCatch;
use vars qw/ @EXPORT_OK /;
@EXPORT_OK = qw/ run_analysis group_variants analyze_variant_location wit_stringify /;
+my $SOLVER_URL = 'http://byzantini.st/cgi-bin/graphcalc.cgi';
+my $unsolved_problems = {};
+
=head1 NAME
Text::Tradition::Analysis - functions for stemma analysis of a tradition
243 => 1,
);
-my $data = run_analysis( $tradition );
+my $data = run_analysis( $tradition, calcdsn => 'dbi:SQLite:dbname=t/data/analysis.db' );
+my $c = $tradition->collation;
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'}},
+ my $gen_bool = $row->{'genealogical'} ? 1 : 0;
+ is( $gen_bool, $expected_genealogical{$row->{'id'}},
"Got correct genealogical flag for row " . $row->{'id'} );
+ # Check that we have the right row with the right groups
+ my $rank = $row->{'id'};
+ foreach my $rdghash ( @{$row->{'readings'}} ) {
+ # Skip 'readings' that aren't really
+ next unless $c->reading( $rdghash->{'readingid'} );
+ # Check the rank
+ is( $c->reading( $rdghash->{'readingid'} )->rank, $rank,
+ "Got correct reading rank" );
+ # Check the witnesses
+ my @realwits = sort $c->reading_witnesses( $rdghash->{'readingid'} );
+ my @sgrp = sort @{$rdghash->{'group'}};
+ is_deeply( \@sgrp, \@realwits, "Reading analyzed with correct groups" );
+ }
}
is( $data->{'variant_count'}, 58, "Got right total variant number" );
# TODO Make something meaningful of conflict count, maybe test other bits
sub run_analysis {
my( $tradition, %opts ) = @_;
my $c = $tradition->collation;
+ my $aclabel = $c->ac_label;
my $stemma_id = $opts{'stemma_id'} || 0;
my @ranks = ref( $opts{'ranks'} ) eq 'ARRAY' ? @{$opts{'ranks'}} : ();
- my @collapse = ref( $opts{'merge_types'} ) eq 'ARRAY' ? @{$opts{'merge_types'}} : ();
+ my $collapse = Set::Scalar->new();
+ if( $opts{'merge_types'} && ref( $opts{'merge_types'} ) eq 'ARRAY' ) {
+ $collapse->insert( @{$opts{'merge_types'}} );
+ } elsif( $opts{'merge_types'} ) {
+ $collapse->insert( $opts{'merge_types'} );
+ }
+
+ # Make sure we have a lookup DB for graph calculation results; this will die
+ # if calcdir or calcdsn isn't passed.
+ my $dir;
+ if( exists $opts{'calcdir'} ) {
+ $dir = delete $opts{'calcdir'}
+ } elsif ( exists $opts{'calcdsn'} ) {
+ $dir = Text::Tradition::Directory->new( dsn => $opts{'calcdsn'} );
+ }
# 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, $_->sigil.$c->ac_label ) if $_->is_layered }
- $tradition->witnesses;
- push( @lacunose, _symmdiff( [ $stemma->witnesses ], \@tradition_wits ) );
+ # Figure out which witnesses we are working with - that is, the ones that
+ # appear both in the stemma and in the tradition. All others are 'lacunose'
+ # for our purposes.
+ my $lacunose = Set::Scalar->new( $stemma->hypotheticals );
+ my $stemma_wits = Set::Scalar->new( $stemma->witnesses );
+ my $tradition_wits = Set::Scalar->new( map { $_->sigil } $tradition->witnesses );
+ $lacunose->insert( $stemma_wits->symmetric_difference( $tradition_wits )->members );
# Find and mark 'common' ranks for exclusion, unless they were
# explicitly specified.
my @groups;
my @use_ranks;
my %lacunae;
+ my $moved = {};
foreach my $rank ( @ranks ) {
- $DB::single = 1 if $rank == 1003;
- my $missing = [ @lacunose ];
- my $rankgroup = group_variants( $tradition, $rank, $missing, \@collapse );
+ my $missing = $lacunose->clone();
+ my $rankgroup = group_variants( $tradition, $rank, $missing, $moved, $collapse );
+ # Filter out any empty rankgroups
+ # (e.g. from the later rank for a transposition)
+ next unless keys %$rankgroup;
+ # Get the graph for this rankgroup
+ my $rankgraph = _graph_for_grouping( $stemma, $rankgroup, $missing, $aclabel );
if( $opts{'exclude_type1'} ) {
# Check to see whether this is a "useful" group.
- my( $rdgs, $grps ) = _useful_variant( $rankgroup,
- $stemma->graph, $c->ac_label );
- next unless @$rdgs;
+ next unless _useful_variant( $rankgroup, $rankgraph, $aclabel );
}
push( @use_ranks, $rank );
- push( @groups, $rankgroup );
+ push( @groups, { grouping => $rankgroup, graph => $rankgraph } );
$lacunae{$rank} = $missing;
}
- # Parse the answer
- my $answer = solve_variants( $stemma, @groups );
+ # Run the solver
+ my $answer;
+ try {
+ $answer = solve_variants( $dir, @groups );
+ } catch ( Text::Tradition::Error $e ) {
+ if( $e->message =~ /IDP/ ) {
+ # Something is wrong with the solver; make the variants table anyway
+ $answer->{'variants'} = [];
+ map { push( @{$answer->{'variants'}}, _init_unsolved( $_, 'IDP error' ) ) }
+ @groups;
+ } else {
+ # Something else is wrong; error out.
+ $e->throw;
+ }
+ }
# Do further analysis on the answer
my $conflict_count = 0;
+ my $reversion_count = 0;
foreach my $idx ( 0 .. $#use_ranks ) {
my $location = $answer->{'variants'}->[$idx];
# Add the rank back in
- $location->{'id'} = $use_ranks[$idx];
- # Add the lacunae back in
- $location->{'missing'} = $lacunae{$use_ranks[$idx]};
+ my $rank = $use_ranks[$idx];
+ $location->{'id'} = $rank;
+ # Note what our lacunae are
my %lmiss;
- map { $lmiss{$_} = 1 } @{$location->{'missing'}};
+ map { $lmiss{$_} = 1 } @{$lacunae{$use_ranks[$idx]}};
+ $location->{'missing'} = [ keys %lmiss ];
+
# Run the extra analysis we need.
- analyze_location( $tradition, $stemma->graph, $location );
+ ## TODO We run through all the variants in this call, so
+ ## why not add the reading data there instead of here below?
+ my $graph = $groups[$idx]->{graph};
+ analyze_location( $tradition, $graph, $location, \%lmiss );
+
+ my @layerwits;
+ # Do the final post-analysis tidying up of the data.
foreach my $rdghash ( @{$location->{'readings'}} ) {
- $conflict_count++
- if exists $rdghash->{'conflict'} && $rdghash->{'conflict'};
- # Add the reading text back in
+ $conflict_count++ if $rdghash->{'is_conflict'};
+ $reversion_count++ if $rdghash->{'is_reverted'};
+ # Add the reading text back in, setting display value as needed
my $rdg = $c->reading( $rdghash->{'readingid'} );
- $rdghash->{'text'} = $rdg ? $rdg->text : $rdghash->{'readingid'};
+ if( $rdg ) {
+ $rdghash->{'text'} = $rdg->text .
+ ( $rdg->rank == $rank ? '' : ' [' . $rdg->rank . ']' );
+ $rdghash->{'is_ungrammatical'} = $rdg->grammar_invalid;
+ $rdghash->{'is_nonsense'} = $rdg->is_nonsense;
+ }
# Remove lacunose witnesses from this reading's list now that the
- # analysis is done
+ # analysis is done
my @realgroup;
- map { push( @realgroup, $_ ) unless $lmiss{$_} } $rdghash->{'group'};
+ map { push( @realgroup, $_ ) unless $lmiss{$_} } @{$rdghash->{'group'}};
$rdghash->{'group'} = \@realgroup;
+ # Note any layered witnesses that appear in this group
+ foreach( @realgroup ) {
+ if( $_ =~ /^(.*)\Q$aclabel\E$/ ) {
+ push( @layerwits, $1 );
+ }
+ }
}
+ $location->{'layerwits'} = \@layerwits if @layerwits;
}
$answer->{'conflict_count'} = $conflict_count;
+ $answer->{'reversion_count'} = $reversion_count;
return $answer;
}
-=head2 group_variants( $tradition, $rank, $lacunose, @merge_relationship_types )
+=head2 group_variants( $tradition, $rank, $lacunose, $transposed, $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.
+relationships in the set $merge_relationship_types as equivalent.
+$lacunose should be a reference to an array, to which the sigla of lacunose
+witnesses at this rank will be appended; $transposed should be a reference
+to a hash, wherein the identities of transposed readings and their
+relatives will be stored.
Returns a hash $group_readings where $rdg is attested by the witnesses listed
in $group_readings->{$rdg}.
# Return group_readings, groups, lacunose
sub group_variants {
- my( $tradition, $rank, $lacunose, $collapse ) = @_;
+ my( $tradition, $rank, $lacunose, $transposed, $collapse ) = @_;
my $c = $tradition->collation;
- my $aclabel = $c->ac_label;
+ my $aclabel = $c->ac_label;
+ my $table = $c->alignment_table;
# Get the alignment table readings
my %readings_at_rank;
- my %is_lacunose; # lookup table for $lacunose
- map { $is_lacunose{$_} = 1 } @$lacunose;
- my @gap_wits;
- foreach my $tablewit ( @{$c->alignment_table->{'alignment'}} ) {
+ my $check_for_gaps = Set::Scalar->new();
+ my %moved_wits;
+ my $has_transposition;
+ foreach my $tablewit ( @{$table->{'alignment'}} ) {
my $rdg = $tablewit->{'tokens'}->[$rank-1];
my $wit = $tablewit->{'witness'};
# Exclude the witness if it is "lacunose" which if we got here
# means "not in the stemma".
- next if $is_lacunose{$wit};
+ next if _is_lacunose( $wit, $lacunose, $aclabel );
+ # Note if the witness is actually in a lacuna
if( $rdg && $rdg->{'t'}->is_lacuna ) {
_add_to_witlist( $wit, $lacunose, $aclabel );
+ # Otherwise the witness either has a positive reading...
} elsif( $rdg ) {
- $readings_at_rank{$rdg->{'t'}->text} = $rdg->{'t'};
+ # If the reading has been counted elsewhere as a transposition, ignore it.
+ if( $transposed->{$rdg->{'t'}->id} ) {
+ # TODO Does this cope with three-way transpositions?
+ map { $moved_wits{$_} = 1 } @{$transposed->{$rdg->{'t'}->id}};
+ next;
+ }
+ # Otherwise, record it...
+ $readings_at_rank{$rdg->{'t'}->id} = $rdg->{'t'};
+ # ...and grab any transpositions, and their relations.
+ my @transp = grep { $_->rank != $rank } $rdg->{'t'}->related_readings();
+ foreach my $trdg ( @transp ) {
+ next if exists $readings_at_rank{$trdg->id};
+ $has_transposition = 1;
+ my @affected_wits = _table_witnesses(
+ $table, $trdg, $lacunose, $aclabel );
+ next unless @affected_wits;
+ map { $moved_wits{$_} = 1 } @affected_wits;
+ $transposed->{$trdg->id} =
+ [ _table_witnesses( $table, $rdg->{'t'}, $lacunose, $aclabel ) ];
+ $readings_at_rank{$trdg->id} = $trdg;
+ }
+ # ...or it is empty, ergo a gap.
} else {
- _add_to_witlist( $wit, \@gap_wits, $aclabel );
+ _add_to_witlist( $wit, $check_for_gaps, $aclabel );
}
}
-
- # Group the readings, collapsing groups by relationship if needed
- my %grouped_readings;
- foreach my $rdg ( sort { $b->witnesses <=> $a->witnesses }
- values %readings_at_rank ) {
+ my $gap_wits = Set::Scalar->new();
+ map { _add_to_witlist( $_, $gap_wits, $aclabel )
+ unless $moved_wits{$_} } $check_for_gaps->members;
+
+ # Group the readings, collapsing groups by relationship if needed.
+ my $grouped_readings = {};
+ foreach my $rdg ( values %readings_at_rank ) {
# Skip readings that have been collapsed into others.
- next if exists $grouped_readings{$rdg->id} && !$grouped_readings{$rdg->id};
- my @wits = $rdg->witnesses;
- if( $collapse ) {
- my $filter = sub { my $r = $_[0]; grep { $_ eq $r->type } @$collapse; };
+ next if exists $grouped_readings->{$rdg->id}
+ && $grouped_readings->{$rdg->id} eq 'COLLAPSE';
+ # Get the witness list, including from readings collapsed into this one.
+ my @wits = _table_witnesses( $table, $rdg, $lacunose, $aclabel );
+ if( $collapse && $collapse->size ) {
+ my $filter = sub { $collapse->has( $_[0]->type ) };
foreach my $other ( $rdg->related_readings( $filter ) ) {
- my @otherwits = $other->witnesses;
+ my @otherwits = _table_witnesses( $table, $other, $lacunose, $aclabel );
push( @wits, @otherwits );
- $grouped_readings{$other->id} = 0;
+ $grouped_readings->{$other->id} = 'COLLAPSE';
}
}
- my @use_wits = grep { !$is_lacunose{$_} } @wits;
- $grouped_readings{$rdg->id} = \@use_wits;
+ $grouped_readings->{$rdg->id} = Set::Scalar->new( @wits );
+ }
+ if( $gap_wits->members ) {
+ $grouped_readings->{'(omitted)'} = $gap_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;
+ map { delete $grouped_readings->{$_} if(
+ $grouped_readings->{$_} eq 'COLLAPSE'
+ || $grouped_readings->{$_}->is_empty ) }
+ keys %$grouped_readings;
+
+ # If something was transposed, check the groups for doubled-up readings
+ if( $has_transposition ) {
+ # print STDERR "Group for rank $rank:\n";
+ # map { print STDERR "\t$_: " . join( ' ' , @{$grouped_readings->{$_}} ) . "\n" }
+ # keys %$grouped_readings;
+ _check_transposed_consistency( $c, $rank, $transposed, $grouped_readings );
+ }
- return \%grouped_readings;
+ # Return the result
+ return $grouped_readings;
}
-=head2 solve_variants( $graph, @groups )
+# Helper function to query the alignment table for all witnesses (a.c. included)
+# that have a given reading at its rank.
+sub _table_witnesses {
+ my( $table, $trdg, $lacunose, $aclabel ) = @_;
+ my $tableidx = $trdg->rank - 1;
+ my $has_reading = Set::Scalar->new();
+ foreach my $row ( @{$table->{'alignment'}} ) {
+ my $wit = $row->{'witness'};
+ next if _is_lacunose( $wit, $lacunose, $aclabel );
+ my $rdg = $row->{'tokens'}->[$tableidx];
+ next unless exists $rdg->{'t'} && defined $rdg->{'t'};
+ _add_to_witlist( $wit, $has_reading, $aclabel )
+ if $rdg->{'t'}->id eq $trdg->id;
+ }
+ return $has_reading->members;
+}
-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.
+# Helper function to see if a witness is lacunose even if we are asking about
+# the a.c. version
+sub _is_lacunose {
+ my ( $wit, $lac, $acstr ) = @_;
+ if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
+ $wit = $1;
+ }
+ return $lac->has( $wit );
+}
+
+# Helper function to ensure that X and X a.c. never appear in the same list.
+sub _add_to_witlist {
+ my( $wit, $list, $acstr ) = @_;
+ if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
+ # Don't add X a.c. if we already have X
+ return if $list->has( $1 );
+ } else {
+ # Delete X a.c. if we are about to add X
+ $list->delete( $wit.$acstr );
+ }
+ $list->insert( $wit );
+}
+
+sub _check_transposed_consistency {
+ my( $c, $rank, $transposed, $groupings ) = @_;
+ my %seen_wits;
+ my %thisrank;
+ # Note which readings are actually at this rank, and which witnesses
+ # belong to which reading.
+ foreach my $rdg ( keys %$groupings ) {
+ my $rdgobj = $c->reading( $rdg );
+ # Count '(omitted)' as a reading at this rank
+ $thisrank{$rdg} = 1 if !$rdgobj || $rdgobj->rank == $rank;
+ map { push( @{$seen_wits{$_}}, $rdg ) } @{$groupings->{$rdg}};
+ }
+ # Our work is done if we have no witness belonging to more than one
+ # reading.
+ my @doubled = grep { scalar @{$seen_wits{$_}} > 1 } keys %seen_wits;
+ return unless @doubled;
+ # If we have a symmetric related transposition, drop the non-rank readings.
+ if( @doubled == scalar keys %seen_wits ) {
+ foreach my $rdg ( keys %$groupings ) {
+ if( !$thisrank{$rdg} ) {
+ my $groupstr = wit_stringify( $groupings->{$rdg} );
+ my ( $matched ) = grep { $groupstr eq wit_stringify( $groupings->{$_} ) }
+ keys %thisrank;
+ delete $groupings->{$rdg};
+ # If we found a group match, assume there is a symmetry happening.
+ # TODO think more about this
+ # print STDERR "*** Deleting symmetric reading $rdg\n";
+ unless( $matched ) {
+ delete $transposed->{$rdg};
+ warn "Found problem in evident symmetry with reading $rdg";
+ }
+ }
+ }
+ # Otherwise 'unhook' the transposed reading(s) that have duplicates.
+ } else {
+ foreach my $dup ( @doubled ) {
+ foreach my $rdg ( @{$seen_wits{$dup}} ) {
+ next if $thisrank{$rdg};
+ next unless exists $groupings->{$rdg};
+ # print STDERR "*** Deleting asymmetric doubled-up reading $rdg\n";
+ delete $groupings->{$rdg};
+ delete $transposed->{$rdg};
+ }
+ }
+ # and put any now-orphaned readings into an 'omitted' reading.
+ foreach my $wit ( keys %seen_wits ) {
+ unless( grep { exists $groupings->{$_} } @{$seen_wits{$wit}} ) {
+ $groupings->{'(omitted)'} = Set::Scalar->new()
+ unless exists $groupings->{'(omitted)'};
+ _add_to_witlist( $wit, $groupings->{'(omitted)'}, $c->ac_label );
+ }
+ }
+ }
+}
+
+# For the given grouping, return its situation graph based on the stemma.
+sub _graph_for_grouping {
+ my( $stemma, $grouping, $lacunose, $aclabel ) = @_;
+ my $acwits = [];
+ my $extant = {};
+ foreach my $gs ( values %$grouping ) {
+ map {
+ if( $_ =~ /^(.*)\Q$aclabel\E$/ ) {
+ push( @$acwits, $1 ) unless $lacunose->has( $1 );
+ } else {
+ $extant->{$_} = 1 unless $lacunose->has( $_ );
+ }
+ } $gs->members;
+ }
+ my $graph;
+ try {
+ # contig contains all extant wits and all hypothetical wits
+ # needed to make up the groups.
+ $graph = $stemma->situation_graph( $extant, $acwits );
+ } catch ( Text::Tradition::Error $e ) {
+ throw( "Could not extend graph with given extant and a.c. witnesses: "
+ . $e->message );
+ } catch {
+ throw( "Could not extend graph with a.c. witnesses @$acwits" );
+ }
+ return $graph;
+}
+
+=head2 solve_variants( $calcdir, @groups )
+
+Looks up the set of groups in the answers provided by 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,
=cut
sub solve_variants {
- my( $stemma, @groups ) = @_;
-
- # Make the json with stemma + groups
- my $groupings = [];
- foreach my $ghash ( @groups ) {
- my @grouping;
- foreach my $k ( keys %$ghash ) {
- push( @grouping, $ghash->{$k} );
+ my( @groups ) = @_;
+
+ # Are we using a local result directory, or did we pass an empty value
+ # for the directory?
+ my $dir;
+ unless( ref( $groups[0] ) eq 'HASH' ) {
+ $dir = shift @groups;
+ }
+
+ ## For each graph/group combo, make a Text::Tradition::Analysis::Result
+ ## object so that we can send it off for IDP lookup.
+ my $variants = [];
+ my $genealogical = 0; # counter
+ # TODO Optimize for unique graph problems
+ my %problems;
+ foreach my $graphproblem ( @groups ) {
+ # Construct the calc result key and look up its answer
+ my $problem = Text::Tradition::Analysis::Result->new(
+ graph => $graphproblem->{'graph'},
+ setlist => [ values %{$graphproblem->{'grouping'}} ] );
+ if( exists $problems{$problem->object_key} ) {
+ $problem = $problems{$problem->object_key};
+ } else {
+ $problems{$problem->object_key} = $problem;
}
- push( @$groupings, \@grouping );
+ $graphproblem->{'object'} = $problem;
}
- ## Witness map is a HACK to get around limitations in node names from IDP
- my $witness_map = {};
- my $json = encode_json( _safe_wit_strings( $stemma, $groupings, $witness_map ) );
-
- # 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;
- my $used_idp;
- if( $resp->is_success ) {
- $answer = _desanitize_names( decode_json( $resp->content ), $witness_map );
- $used_idp = 1;
+
+ my %results;
+ if( $dir ) {
+ my $scope = $dir->new_scope;
+ map { $results{$_} = $dir->lookup( $_ ) || $problems{$_} } keys %problems;
} else {
- # Fall back to the old method.
- warn "IDP solver returned " . $resp->status_line . " / " . $resp->content
- . "; falling back to perl method";
- $answer = perl_solver( $stemma, @$groupings );
+ my $json = JSON->new->allow_blessed->convert_blessed->utf8->encode(
+ [ values %problems ] );
+ # Send it off and get the result
+ # print STDERR "Sending request: " . decode_utf8( $json ) . "\n";
+ 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 );
+ throw( "Unexpected answer from IDP: $answer" ) unless ref( $answer ) eq 'ARRAY';
+ } else {
+ throw( "IDP solver returned " . $resp->status_line . " / " . $resp->content
+ . "; cannot run graph analysis" );
+ }
+ # One more sanity check
+ throw( "Something went wrong with answer symmetricity" )
+ unless keys( %problems ) == @$answer;
+ # Convert the results
+ foreach my $a ( @$answer ) {
+ my $r = Text::Tradition::Analysis::Result->new( $a );
+ $results{$r->object_key} = $r;
+ }
}
- # 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]};
- if( $result ) {
- $genealogical++;
- # Prune the calculated groups, in case the IDP solver failed to.
- if( $used_idp ) {
- my @pruned_groups;
- foreach my $cg ( @$calc_groups ) {
- my @pg = _prune_group( $cg, $stemma );
- push( @pruned_groups, \@pg );
- }
- $calc_groups = \@pruned_groups;
- }
- }
- my $input_group = $groups[$idx];
- foreach my $k ( keys %$input_group ) {
- my $cg = shift @$calc_groups;
- $input_group->{$k} = $cg;
+ # We now have a single JSON-encoded Result object per problem sent. Fold its
+ # answers into our variant info structure.
+ foreach my $graphproblem ( @groups ) {
+ my $result = $results{$graphproblem->{'object'}->object_key}
+ || $graphproblem->{'object'};
+
+ # Initialize the result structure for this graph problem
+ my $vstruct;
+ if( $result->status eq 'OK' ) {
+ $vstruct = { readings => [] };
+ push( @$variants, $vstruct );
+ } else {
+ push( @$variants, _init_unsolved( $graphproblem, $result->status ) );
+ next;
}
- my $vstruct = {
- 'genealogical' => $result,
- 'readings' => [],
- };
- foreach my $k ( sort { @{$input_group->{$b}} <=> @{$input_group->{$a}} }
- keys %$input_group ) {
- push( @{$vstruct->{'readings'}},
- { 'readingid' => $k, 'group' => $input_group->{$k}} );
+
+ # 1. Did the group evaluate as genealogical?
+ $vstruct->{genealogical} = $result->is_genealogical;
+ $genealogical++ if $result->is_genealogical;
+
+ # 2. What are the calculated minimum groupings for each variant loc?
+ foreach my $rid ( keys %{$graphproblem->{grouping}} ) {
+ my $inputset = $graphproblem->{grouping}->{$rid};
+ my $minset = $result->minimum_grouping_for( $inputset );
+ push( @{$vstruct->{readings}}, { readingid => $rid, group => $minset } );
}
- push( @$variants, $vstruct );
+
+ # 3. What are the sources and classes calculated for each witness?
+ $vstruct->{witcopy_types} = { $result->classes };
+ $vstruct->{reading_roots} = {};
+ map { $vstruct->{reading_roots}->{$_} = 1 } $result->sources;
+
}
return { 'variants' => $variants,
'genealogical_count' => $genealogical };
}
-#### HACKERY to cope with IDP's limited idea of what a node name looks like ###
-
-sub _safe_wit_strings {
- my( $stemma, $groupings, $witness_map ) = @_;
- my $safegraph = Graph->new();
- # Convert the graph to a safe representation and store the conversion.
- foreach my $n ( $stemma->graph->vertices ) {
- my $sn = _safe_witstr( $n );
- warn "Ambiguous stringification $sn for $n and " . $witness_map->{$sn}
- if exists $witness_map->{$sn};
- $witness_map->{$sn} = $n;
- $safegraph->add_vertex( $sn );
- $safegraph->set_vertex_attributes( $sn,
- $stemma->graph->get_vertex_attributes( $n ) );
- }
- foreach my $e ( $stemma->graph->edges ) {
- my @safe_e = ( _safe_witstr( $e->[0] ), _safe_witstr( $e->[1] ) );
- $safegraph->add_edge( @safe_e );
- }
- my $safe_stemma = Text::Tradition::Stemma->new(
- 'collation' => $stemma->collation, 'graph' => $safegraph );
-
- # Now convert the witness groupings to a safe representation.
- my $safe_groupings = [];
- foreach my $grouping ( @$groupings ) {
- my $safe_grouping = [];
- foreach my $group ( @$grouping ) {
- my $safe_group = [];
- foreach my $n ( @$group ) {
- my $sn = _safe_witstr( $n );
- warn "Ambiguous stringification $sn for $n and " . $witness_map->{$sn}
- if exists $witness_map->{$sn} && $witness_map->{$sn} ne $n;
- $witness_map->{$sn} = $n;
- push( @$safe_group, $sn );
- }
- push( @$safe_grouping, $safe_group );
- }
- push( @$safe_groupings, $safe_grouping );
- }
-
- # Return it all in the struct we expect. We have stored the reductions
- # in the $witness_map that we were passed.
- return { 'graph' => $safe_stemma->editable( ' ' ), 'groupings' => $safe_groupings };
-}
-
-sub _safe_witstr {
- my $witstr = shift;
- $witstr =~ s/\s+/_/g;
- $witstr =~ s/[^\w\d-]//g;
- return $witstr;
-}
-
-sub _desanitize_names {
- my( $jsonstruct, $witness_map ) = @_;
- my $result = [];
- foreach my $grouping ( @$jsonstruct ) {
- my $real_grouping = [];
- foreach my $element ( @$grouping ) {
- if( ref( $element ) eq 'ARRAY' ) {
- # it's the groupset.
- my $real_groupset = [];
- foreach my $group ( @$element ) {
- my $real_group = [];
- foreach my $n ( @$group ) {
- my $rn = $witness_map->{$n};
- push( @$real_group, $rn );
- }
- push( @$real_groupset, $real_group );
- }
- push( @$real_grouping, $real_groupset );
- } else {
- # It is the boolean, not actually a group.
- push( @$real_grouping, $element );
- }
- }
- push( @$result, $real_grouping );
+sub _init_unsolved {
+ my( $graphproblem, $status ) = @_;
+ my $vstruct = { 'readings' => [] };
+ $vstruct->{'unsolved'} = $status;
+ foreach my $rid ( keys %{$graphproblem->{grouping}} ) {
+ push( @{$vstruct->{readings}}, { readingid => $rid,
+ group => [ $graphproblem->{grouping}->{$rid}->members ] } );
}
- return $result;
+ return $vstruct;
}
-### END HACKERY ###
-
=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.
+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 ) = @_;
+ my ( $tradition, $graph, $variant_row, $lacunose ) = @_;
+ my $c = $tradition->collation;
+
+ if( exists $variant_row->{'unsolved'} ) {
+ return;
+ }
+ my $reading_roots = delete $variant_row->{'reading_roots'};
+ my $classinfo = delete $variant_row->{'witcopy_types'};
# Make a hash of all known node memberships, and make the subgraphs.
my $contig = {};
- my $reading_roots = {};
my $subgraph = {};
+ my $acstr = $c->ac_label;
+ my @acwits;
+
+ # Note which witnesses positively belong to which group. This information
+ # comes ultimately from the IDP solver.
+ # Also make a note of the reading's roots.
foreach my $rdghash ( @{$variant_row->{'readings'}} ) {
my $rid = $rdghash->{'readingid'};
- map { $contig->{$_} = $rid } @{$rdghash->{'group'}};
-
- # 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;
+ my @roots;
+ foreach my $wit ( @{$rdghash->{'group'}} ) {
+ $contig->{$wit} = $rid;
+ if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
+ push( @acwits, $1 );
+ }
+ if( exists $reading_roots->{$wit} && $reading_roots->{$wit} ) {
+ push( @roots, $wit );
+ }
+ }
+ $rdghash->{'independent_occurrence'} = \@roots;
}
-
+
# 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'}} ) {
- # 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};
+ my $rdg = $c->reading( $rid );
+ my @roots = @{$rdghash->{'independent_occurrence'}};
+ my @reversions;
+ if( $classinfo ) {
+ @reversions = grep { $classinfo->{$_} eq 'revert' }
+ $rdghash->{'group'}->members;
+ $rdghash->{'reversions'} = \@reversions;
+ }
+ my @group = @{$rdghash->{'group'}};
# Start figuring things out.
- my @roots = $part->predecessorless_vertices;
- $rdghash->{'independent_occurrence'} = scalar @roots;
- $rdghash->{'followed'} = scalar( $part->vertices ) - scalar( @roots );
+ $rdghash->{'followed'} = scalar( @group )
+ - ( scalar( @roots ) + scalar( @reversions ) );
# Find the parent readings, if any, of this reading.
- my %rdgparents;
- foreach my $wit ( @roots ) {
- # Look in the main stemma to find this witness's extant or known-reading
- # immediate ancestor(s), and look up the reading that each ancestor olds.
- my @check = $graph->predecessors( $wit );
- while( @check ) {
- my @next;
- foreach my $wparent( @check ) {
- my $preading = $contig->{$wparent};
- if( $preading ) {
- $rdgparents{$preading} = 1;
- } else {
- push( @next, $graph->predecessors( $wparent ) );
- }
- }
- @check = @next;
- }
+ my $sourceparents = _find_reading_parents( $rid, $graph, $contig, @roots );
+ # Work out relationships between readings and their non-followed parent.
+ _resolve_parent_relationships( $c, $rid, $rdg, $sourceparents );
+ $rdghash->{'source_parents'} = $sourceparents;
+
+ if( @reversions ) {
+ my $revparents = _find_reading_parents( $rid, $graph, $contig, @reversions );
+ _resolve_parent_relationships( $c, $rid, $rdg, $revparents );
+ $rdghash->{'reversion_parents'} = $revparents;
}
- $rdghash->{'reading_parents'} = [ keys %rdgparents ];
# Find the number of times this reading was altered, and the number of
# times we're not sure.
my( %nofollow, %unknownfollow );
- foreach my $wit ( $part->vertices ) {
+ foreach my $wit ( @{$rdghash->{'group'}} ) {
foreach my $wchild ( $graph->successors( $wit ) ) {
- next if $part->has_vertex( $wchild );
- if( $reading_roots->{$wchild} && $contig->{$wchild} ) {
+ if( $reading_roots->{$wchild} && $contig->{$wchild}
+ && $contig->{$wchild} ne $rid ) {
# It definitely changed here.
$nofollow{$wchild} = 1;
} elsif( !($contig->{$wchild}) ) {
# The child is a hypothetical node not definitely in
# any group. Answer is unknown.
$unknownfollow{$wchild} = 1;
- } # else it's a non-root node in a known group, and therefore
- # is presumed to have its reading from its group, not this link.
+ } # else it is either in our group, or it is a non-root node in a
+ # known group and therefore is presumed to have its reading from
+ # its group, not this link.
}
}
$rdghash->{'not_followed'} = keys %nofollow;
# Now say whether this reading represents a conflict.
unless( $variant_row->{'genealogical'} ) {
- $rdghash->{'conflict'} = @roots != 1;
+ $rdghash->{'is_conflict'} = @roots != 1;
+ $rdghash->{'is_reverted'} = scalar @reversions;
}
}
}
-
-=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( $stemma, @groups ) = @_;
- my $graph = $stemma->graph;
- my @answer;
- foreach my $g ( @groups ) {
- push( @answer, _solve_variant_location( $graph, $g ) );
+sub _find_reading_parents {
+ my( $rid, $graph, $contig, @list ) = @_;
+ my $parenthash = {};
+ foreach my $wit ( @list ) {
+ # Look in the stemma graph to find this witness's extant or known-reading
+ # immediate ancestor(s), and look up the reading that each ancestor holds.
+ my @check = $graph->predecessors( $wit );
+ while( @check ) {
+ my @next;
+ foreach my $wparent( @check ) {
+ my $preading = $contig->{$wparent};
+ if( $preading && $preading ne $rid ) {
+ $parenthash->{$preading} = 1;
+ } else {
+ push( @next, $graph->predecessors( $wparent ) );
+ }
+ }
+ @check = @next;
+ }
}
- return \@answer;
+ return $parenthash;
}
-sub _solve_variant_location {
- my( $graph, $groups ) = @_;
- # Now do the work.
- my $contig = {};
- my $subgraph = {};
- my $is_conflicted;
- my $conflict = {};
-
- # 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 } $graph->vertices );
-
- # Now look to see if our group is connected.
- if( @$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 = $g->[0];
- @group_roots = ( $wit );
- foreach my $v ( $part->vertices ) {
- $part->delete_vertex( $v ) unless $v eq $wit;
- }
- }
-
- if( @group_roots > 1 ) {
- $conflict->{$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?";
- }
- }
-
-
- # Save the relevant subgraph.
- $subgraph->{$gst} = $part;
- }
-
- # 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.
+sub _resolve_parent_relationships {
+ my( $c, $rid, $rdg, $rdgparents ) = @_;
+ foreach my $p ( keys %$rdgparents ) {
+ # Resolve the relationship of the parent to the reading, and
+ # save it in our hash.
+ my $pobj = $c->reading( $p );
+ my $prep = $pobj ? $pobj->id . ' (' . $pobj->text . ')' : $p;
+ my $phash = { 'label' => $prep };
+ if( $pobj ) {
+ my $rel = $c->get_relationship( $p, $rid );
+ if( $rel ) {
+ _add_to_hash( $rel, $phash );
+ } elsif( $rdg ) {
+ # First check for a transposed relationship
+ if( $rdg->rank != $pobj->rank ) {
+ foreach my $ti ( $rdg->related_readings( 'transposition' ) ) {
+ next unless $ti->text eq $rdg->text;
+ $rel = $c->get_relationship( $ti, $pobj );
+ if( $rel ) {
+ _add_to_hash( $rel, $phash, 1 );
+ last;
+ }
+ }
+ unless( $rel ) {
+ foreach my $ti ( $pobj->related_readings( 'transposition' ) ) {
+ next unless $ti->text eq $pobj->text;
+ $rel = $c->get_relationship( $ti, $rdg );
+ if( $rel ) {
+ _add_to_hash( $rel, $phash, 1 );
+ last;
+ }
+ }
+ }
+ }
+ unless( $rel ) {
+ # and then check for sheer word similarity.
+ my $rtext = $rdg->text;
+ my $ptext = $pobj->text;
+ if( similar( $rtext, $ptext ) ) {
+ # say STDERR "Words $rtext and $ptext judged similar";
+ $phash->{relation} = { type => 'wordsimilar' };
+ }
+ }
} else {
- push( @resolve, $wit );
+ $phash->{relation} = { type => 'deletion' };
}
- } else {
- my $gst = pop @{$contig->{$wit}};
- $contig->{$wit} = $gst || '';
+ # Get the attributes of the parent object while we are here
+ $phash->{'text'} = $pobj->text if $pobj;
+ $phash->{'is_nonsense'} = $pobj->is_nonsense;
+ $phash->{'is_ungrammatical'} = $pobj->grammar_invalid;
+ } elsif( $p eq '(omitted)' ) {
+ $phash->{relation} = { type => 'addition' };
}
+ # Save it
+ $rdgparents->{$p} = $phash;
}
-
- 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!
- $conflict->{$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!
- $conflict->{$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
-
- my $is_genealogical = keys %$conflict ? JSON::false : JSON::true;
- my $variant_row = [ [], $is_genealogical ];
- # Fill in the groupings from $contig.
- foreach my $g ( @$groups ) {
- my $gst = wit_stringify( $g );
- my @realgroup = grep { $contig->{$_} eq $gst } keys %$contig;
- push( @{$variant_row->[0]}, \@realgroup );
- }
- return $variant_row;
}
-sub _prune_group {
- my( $group, $stemma ) = @_;
- # Get these into a form prune_subtree will recognize. Make a "contighash"
- my $hypohash = {};
- map { $hypohash->{$_} = 1 } @$group;
- # ...with reference values for hypotheticals.
- map { $hypohash->{$_} = [] } $stemma->hypotheticals;
- # Make our subgraph
- my $subgraph = $stemma->graph->copy;
- map { $subgraph->delete_vertex( $_ ) unless exists $hypohash->{$_} }
- $subgraph->vertices;
- # ...and find the root.
- my( $root ) = $subgraph->predecessorless_vertices;
- # Now prune and return the remaining vertices.
- _prune_subtree( $subgraph, $root, $hypohash );
- return $subgraph->vertices;
+sub _add_to_hash {
+ my( $rel, $phash, $is_transposed ) = @_;
+ $phash->{relation} = { type => $rel->type };
+ $phash->{relation}->{transposed} = 1 if $is_transposed;
+ $phash->{relation}->{annotation} = $rel->annotation
+ if $rel->has_annotation;
}
-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 first child that has no other predecessors.
- while( $tree->successors( $root ) == 1 && ref $contighash->{$root} ) {
- my @nextroot = $tree->successors( $root );
- $tree->delete_vertex( $root );
- ( $root ) = grep { $tree->is_predecessorless_vertex( $_ ) } @nextroot;
- }
- # The tree has been modified in place, but we need to know the new root.
- $root = undef unless $root && $tree->has_vertex( $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 {
- my( $arr, $wit, $acstr ) = @_;
- my $skip;
- if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
- my $real = $1;
- $skip = grep { $_ =~ /^\Q$real\E$/ } @$arr;
- }
- push( @$arr, $wit ) unless $skip;
+=head2 similar( $word1, $word2 )
+
+Use Algorithm::Diff to get a sense of how close the words are to each other.
+This will hopefully handle substitutions a bit more nicely than Levenshtein.
+
+=cut
+
+#!/usr/bin/env perl
+
+sub similar {
+ my( $word1, $word2 ) = sort { length($a) <=> length($b) } @_;
+ my @let1 = split( '', lc( $word1 ) );
+ my @let2 = split( '', lc( $word2 ) );
+ my $diff = Algorithm::Diff->new( \@let1, \@let2 );
+ my $mag = 0;
+ while( $diff->Next ) {
+ if( $diff->Same ) {
+ # Take off points for longer strings
+ my $cs = $diff->Range(1) - 2;
+ $cs = 0 if $cs < 0;
+ $mag -= $cs;
+ } elsif( !$diff->Items(1) ) {
+ $mag += $diff->Range(2);
+ } elsif( !$diff->Items(2) ) {
+ $mag += $diff->Range(1);
+ } else {
+ # Split the difference for substitutions
+ my $c1 = $diff->Range(1) || 1;
+ my $c2 = $diff->Range(2) || 1;
+ my $cd = ( $c1 + $c2 ) / 2;
+ $mag += $cd;
+ }
+ }
+ return ( $mag <= length( $word1 ) / 2 );
}
sub _useful_variant {
- my( $group_readings, $graph, $acstr ) = @_;
-
- # TODO Decide what to do with AC witnesses
+ my( $rankgroup, $rankgraph, $acstr ) = @_;
# Sort by group size and return
my $is_useful = 0;
- my( @readings, @groups ); # The sorted groups for our answer.
- foreach my $rdg ( sort { @{$group_readings->{$b}} <=> @{$group_readings->{$a}} }
- keys %$group_readings ) {
- push( @readings, $rdg );
- push( @groups, $group_readings->{$rdg} );
- if( @{$group_readings->{$rdg}} > 1 ) {
+ foreach my $rdg ( keys %$rankgroup ) {
+ my @wits = $rankgroup->{$rdg}->members;
+ if( @wits > 1 ) {
$is_useful++;
} else {
- my( $wit ) = @{$group_readings->{$rdg}};
- $wit =~ s/^(.*)\Q$acstr\E$/$1/;
- $is_useful++ unless( $graph->is_sink_vertex( $wit ) );
+ $is_useful++ unless( $rankgraph->is_sink_vertex( $wits[0] )
+ || $wits[0] =~ /\Q$acstr\E$/ );
}
}
- if( $is_useful > 1 ) {
- return( \@readings, \@groups );
- } else {
- return( [], [] );
- }
+ return $is_useful > 1;
}
=head2 wit_stringify( $groups )
return join( ' / ', @gst );
}
-# Helper function to ensure that X and X a.c. never appear in the same list.
-sub _add_to_witlist {
- my( $wit, $list, $acstr ) = @_;
- my %inlist;
- my $idx = 0;
- map { $inlist{$_} = $idx++ } @$list;
- if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
- my $acwit = $1;
- unless( exists $inlist{$acwit} ) {
- push( @$list, $acwit.$acstr );
- }
- } else {
- if( exists( $inlist{$wit.$acstr} ) ) {
- # Replace the a.c. version with the main witness
- my $i = $inlist{$wit.$acstr};
- $list->[$i] = $wit;
- } else {
- push( @$list, $wit );
- }
- }
-}
+1;
-sub _symmdiff {
- my( $lista, $listb ) = @_;
- my %union;
- my %scalars;
- map { $union{$_} = 1; $scalars{$_} = $_ } @$lista;
- map { $union{$_} += 1; $scalars{$_} = $_ } @$listb;
- my @set = grep { $union{$_} == 1 } keys %union;
- return map { $scalars{$_} } @set;
+sub throw {
+ Text::Tradition::Error->throw(
+ 'ident' => 'Analysis error',
+ 'message' => $_[0],
+ );
}
-1;
-
=head1 LICENSE
This package is free software and is provided "as is" without express