use strict;
use warnings;
+use Algorithm::Diff; # for word similarity measure
use Benchmark;
use Encode qw/ encode_utf8 /;
use Exporter 'import';
use LWP::UserAgent;
use Text::Tradition;
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';
+
+
=head1 NAME
Text::Tradition::Analysis - functions for stemma analysis of a tradition
my @groups;
my @use_ranks;
my %lacunae;
+ my $moved = {};
foreach my $rank ( @ranks ) {
my $missing = [ @lacunose ];
- my $rankgroup = group_variants( $tradition, $rank, $missing, \@collapse );
+ 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;
if( $opts{'exclude_type1'} ) {
# Check to see whether this is a "useful" group.
my( $rdgs, $grps ) = _useful_variant( $rankgroup,
foreach my $idx ( 0 .. $#use_ranks ) {
my $location = $answer->{'variants'}->[$idx];
# Add the rank back in
- $location->{'id'} = $use_ranks[$idx];
+ my $rank = $use_ranks[$idx];
+ $location->{'id'} = $rank;
# Note what our lacunae are
my %lmiss;
map { $lmiss{$_} = 1 } @{$lacunae{$use_ranks[$idx]}};
- # Run through the reading groups and add as 'lacunae' any redundant
- # a.c. witnesses (yes, we have to do this before the analysis, thus
- # identical loops before and after. Boo.)
- # TODO Consider making these callbacks to analyze_location
- foreach my $rdghash ( @{$location->{'readings'}} ) {
- my %rwits;
- map { $rwits{$_} = 1 } @{$rdghash->{'group'}};
- foreach my $rw ( keys %rwits ) {
- if( $rw =~ /^(.*)\Q$aclabel\E$/ ) {
- if( exists $rwits{$1} ) {
- $lmiss{$rw} = 1;
- delete $rwits{$rw};
- }
- }
- }
- $rdghash->{'group'} = [ keys %rwits ];
- }
$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?
+ analyze_location( $tradition, $stemma, $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
+ # 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
my @realgroup;
map { push( @realgroup, $_ ) unless $lmiss{$_} } @{$rdghash->{'group'}};
$rdghash->{'group'} = \@realgroup;
- # TODO Record hypotheticals used to create group, if we end up
- # needing it
+ # 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;
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.
+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 $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 %is_lacunose; # lookup table for witnesses not in stemma
+ map { $is_lacunose{$_} = 1; $is_lacunose{$_.$aclabel} = 1 } @$lacunose;
+ my @check_for_gaps;
+ 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};
+ # 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, \%is_lacunose, $aclabel );
+ next unless @affected_wits;
+ map { $moved_wits{$_} = 1 } @affected_wits;
+ $transposed->{$trdg->id} =
+ [ _table_witnesses( $table, $rdg->{'t'}, \%is_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 );
}
}
-
+ my @gap_wits;
+ map { _add_to_witlist( $_, \@gap_wits, $aclabel )
+ unless $moved_wits{$_} } @check_for_gaps;
# Group the readings, collapsing groups by relationship if needed
- my %grouped_readings;
+ 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};
+ 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 = $rdg->witnesses;
- if( $collapse ) {
+ my @wits = _table_witnesses( $table, $rdg, \%is_lacunose, $aclabel );
+ if( $collapse && @$collapse ) {
my $filter = sub { my $r = $_[0]; grep { $_ eq $r->type } @$collapse; };
foreach my $other ( $rdg->related_readings( $filter ) ) {
- my @otherwits = $other->witnesses;
+ my @otherwits = _table_witnesses(
+ $table, $other, \%is_lacunose, $aclabel );
push( @wits, @otherwits );
- $grouped_readings{$other->id} = 0;
+ $grouped_readings->{$other->id} = 'COLLAPSE';
}
}
- # Filter the group to those witnesses in the stemma
- my @use_wits;
- foreach my $wit ( @wits ) {
- next if $is_lacunose{$wit};
- push( @use_wits, $wit );
- }
- $grouped_readings{$rdg->id} = \@use_wits;
+ $grouped_readings->{$rdg->id} = \@wits;
}
- $grouped_readings{'(omitted)'} = \@gap_wits if @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
+ map { delete $grouped_readings->{$_} if $grouped_readings->{$_} eq 'COLLAPSE' }
+ keys %$grouped_readings
if $collapse;
+
+ # 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 the result
- return \%grouped_readings;
+ return $grouped_readings;
+}
+
+# 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;
+ foreach my $row ( @{$table->{'alignment'}} ) {
+ my $wit = $row->{'witness'};
+ next if $lacunose->{$wit};
+ 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;
}
# Helper function to ensure that X and X a.c. never appear in the same list.
}
}
+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)'} = [] unless exists $groupings->{'(omitted)'};
+ _add_to_witlist( $wit, $groupings->{'(omitted)'}, $c->ac_label );
+ }
+ }
+ }
+}
+
=head2 solve_variants( $graph, @groups )
Sends the set of groups to the external graph solver service and returns
sub solve_variants {
my( $stemma, @groups ) = @_;
- my $aclabel = $stemma->collation->ac_label;
# Filter the groups down to distinct groups, and work out what graph
# should be used in the calculation of each group. We want to send each
# distinct problem to the solver only once.
# We need a whole bunch of lookup tables for this.
- my $index_groupkeys = {}; # Save the order of readings
- my $group_indices = {}; # Save the indices that have a given grouping
- my $graph_problems = {}; # Save the groupings for the given graph
+ my( $index_groupkeys, $group_indices, $graph_problems ) = _prepare_groups( @_ );
- foreach my $idx ( 0..$#groups ) {
- my $ghash = $groups[$idx];
- my @grouping;
- # Sort the groupings from big to little, and scan for a.c. witnesses
- # that would need an extended graph.
- my @acwits; # note which AC witnesses crop up at this rank
- my @idxkeys = sort { scalar @{$ghash->{$b}} <=> scalar @{$ghash->{$a}} }
- keys %$ghash;
- foreach my $rdg ( @idxkeys ) {
- my @sg = sort @{$ghash->{$rdg}};
- push( @acwits, grep { $_ =~ /\Q$aclabel\E$/ } @sg );
- push( @grouping, \@sg );
- }
- # Save the reading order
- $index_groupkeys->{$idx} = \@idxkeys;
-
- # Now associate the distinct group with this index
- my $gstr = wit_stringify( \@grouping );
- push( @{$group_indices->{$gstr}}, $idx );
-
- # Finally, add the group to the list to be calculated for this graph.
- map { s/\Q$aclabel\E$// } @acwits;
- my $graph = $stemma->extend_graph( \@acwits );
- unless( exists $graph_problems->{"$graph"} ) {
- $graph_problems->{"$graph"} = { 'object' => $graph, 'groups' => [] };
- }
- push( @{$graph_problems->{"$graph"}->{'groups'}}, \@grouping );
- }
-
## For each distinct graph, send its groups to the solver.
- $DB::single = 1;
- my $solver_url = 'http://byzantini.st/cgi-bin/graphcalc.cgi';
my $ua = LWP::UserAgent->new();
## Witness map is a HACK to get around limitations in node names from IDP
my $witness_map = {};
my $genealogical = 0;
foreach my $graphkey ( keys %$graph_problems ) {
my $graph = $graph_problems->{$graphkey}->{'object'};
- my $groupings = $graph_problems->{$graphkey}->{'groups'};
- my $json = encode_json( _safe_wit_strings( $graph, $stemma->collation,
- $groupings, $witness_map ) );
+ my $groupings = [ values %{$graph_problems->{$graphkey}->{'groups'}} ];
+ my $req = _safe_wit_strings( $graph, $stemma->collation,
+ $groupings, $witness_map );
+ $req->{'command'} = 'findGroupings';
+ my $json = encode_json( $req );
# Send it off and get the result
- my $resp = $ua->post( $solver_url, 'Content-Type' => 'application/json',
+ # print STDERR "Sending request: " . to_json( $req ) . "\n";
+ 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;
} else {
# Fall back to the old method.
- warn "IDP solver returned " . $resp->status_line . " / " . $resp->content
- . "; falling back to perl method";
- $answer = perl_solver( $graph, @$groupings );
+ die "IDP solver returned " . $resp->status_line . " / " . $resp->content
+ . "; cannot run graph analysis";
+ }
+
+ ## If IDP worked, asked it the other two questions for this dataset.
+ my $more_eval = {};
+ foreach my $test ( qw/ findSources findClasses / ) {
+ $req->{'command'} = $test;
+ $json = encode_json( $req );
+ $resp = $ua->post( $SOLVER_URL, 'Content-Type' => 'application/json',
+ 'Content' => $json );
+ if( $resp->is_success ) {
+ $more_eval->{$test} = _desanitize_names(
+ decode_json( $resp->content ), $witness_map );
+ } else {
+ warn "IDP solver for $test returned " . $resp->status_line .
+ " / " . $resp->content;
+ # TODO arrange fallback
+ }
}
+
## The answer is the evaluated groupings, plus a boolean for whether
## they were genealogical. Reconstruct our original groups.
foreach my $gidx ( 0 .. $#{$groupings} ) {
my( $calc_groups, $result ) = @{$answer->[$gidx]};
- 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 ) {
- # This is a little wasteful but the path of least
- # resistance. Send both the stemma, which knows what
- # its hypotheticals are, and the actual graph used.
- my @pg = _prune_group( $cg, $stemma, $graph );
- push( @pruned_groups, \@pg );
- }
- $calc_groups = \@pruned_groups;
+ # Keep track of the total # of genealogical readings
+ $genealogical++ if $result;
+
+ my( $sources, $classes );
+ # Use the expanded groups from findSources if that got calculated.
+ if( exists( $more_eval->{'findSources'} ) ) {
+ ( $calc_groups, $sources ) = @{$more_eval->{'findSources'}->[$gidx]};
+ }
+ # Use the (same) expanded groups from findClasses if that got calculated
+ # and is relevant.
+ if( exists( $more_eval->{'findClasses'} ) && !$result ) {
+ ( $calc_groups, $classes ) = @{$more_eval->{'findClasses'}->[$gidx]};
+ }
+
+ # Prune the calculated groups, in case the IDP solver failed to.
+ if( $sources || $result ) {
+ my @pruned_groups;
+ my @pruned_roots;
+ foreach my $cg ( @$calc_groups ) {
+ my( $pg, $pr ) = _prune_group( $cg, $graph );
+ push( @pruned_groups, $pg );
+ push( @pruned_roots, @$pr );
}
+ $calc_groups = \@pruned_groups;
+ say STDERR "Pruned roots from @$sources to @pruned_roots"
+ unless wit_stringify( [ sort @$sources ] )
+ eq wit_stringify( [ sort @pruned_roots ] );
+ $sources = \@pruned_roots;
+ }
+
+ # Convert the source list into a lookup hash
+ my $roots = {};
+ map { $roots->{$_} = 1 } @$sources;
+ # Convert the class list into a lookup hash
+ if( $classes ) {
+ $classes = _invert_hash( $classes );
}
+
# Retrieve the key for the original group that went to the solver
my $input_group = wit_stringify( $groupings->[$gidx] );
+
+ # Make the variant hash for each location that had this particular
+ # grouping on this particular stemma situation
foreach my $oidx ( @{$group_indices->{$input_group}} ) {
my @readings = @{$index_groupkeys->{$oidx}};
my $vstruct = {
{ 'readingid' => $readings[$ridx],
'group' => $calc_groups->[$ridx] } );
}
+ $vstruct->{'reading_roots'} = $roots if $roots;
+ $vstruct->{'reading_types'} = $classes if $classes;
$variants->[$oidx] = $vstruct;
}
}
'genealogical_count' => $genealogical };
}
+sub _prepare_groups {
+ my( $stemma, @groups ) = @_;
+ my $aclabel = $stemma->collation->ac_label;
+
+ my $index_groupkeys = {}; # Save the order of readings
+ my $group_indices = {}; # Save the indices that have a given grouping
+ my $graph_problems = {}; # Save the groupings for the given graph
+
+ foreach my $idx ( 0..$#groups ) {
+ my $ghash = $groups[$idx];
+ my @grouping;
+ # Sort the groupings from big to little, and scan for a.c. witnesses
+ # that would need an extended graph.
+ my @acwits; # note which AC witnesses crop up at this rank
+ my $extant; # note which witnesses crop up at this rank full stop
+ my @idxkeys = sort { scalar @{$ghash->{$b}} <=> scalar @{$ghash->{$a}} }
+ keys %$ghash;
+ foreach my $rdg ( @idxkeys ) {
+ my @sg = sort @{$ghash->{$rdg}};
+ push( @acwits, grep { $_ =~ /\Q$aclabel\E$/ } @sg );
+ map { $extant->{$_} = 1 } @sg;
+ push( @grouping, \@sg );
+ }
+ # Save the reading order
+ $index_groupkeys->{$idx} = \@idxkeys;
+
+ # Now associate the distinct group with this index
+ my $gstr = wit_stringify( \@grouping );
+ push( @{$group_indices->{$gstr}}, $idx );
+
+ # Finally, add the group to the list to be calculated for this graph.
+ map { s/\Q$aclabel\E$// } @acwits;
+ my $graph;
+ ## TODO When we get rid of the safe_wit_strings HACK we should also
+ ## be able to save the graph here as a dotstring rather than as an
+ ## object, thus simplifying life enormously.
+ try {
+ $graph = $stemma->situation_graph( $extant, \@acwits );
+ } catch {
+ $DB::single = 1;
+ die "Unable to extend graph with @acwits";
+ }
+ my $graphkey = "$graph || " . wit_stringify( [ sort keys %$extant ] );
+ unless( exists $graph_problems->{$graphkey} ) {
+ $graph_problems->{$graphkey} = { 'object' => $graph, 'groups' => {} };
+ }
+ $graph_problems->{$graphkey}->{'groups'}->{wit_stringify( \@grouping )} = \@grouping;
+ }
+ say STDERR "Created " . scalar( keys %$graph_problems ). " distinct graph(s)";
+ return( $index_groupkeys, $group_indices, $graph_problems );
+}
+
#### HACKERY to cope with IDP's limited idea of what a node name looks like ###
sub _safe_wit_strings {
my( $graph, $c, $groupings, $witness_map ) = @_;
- # Parse the graph we were given into a stemma.
- my $safegraph = Graph->new();
# Convert the graph to a safe representation and store the conversion.
+ my $safegraph = Graph->new();
foreach my $n ( $graph->vertices ) {
my $sn = _safe_witstr( $n );
if( exists $witness_map->{$sn} ) {
my @safe_e = ( _safe_witstr( $e->[0] ), _safe_witstr( $e->[1] ) );
$safegraph->add_edge( @safe_e );
}
- my $safe_stemma = Text::Tradition::Stemma->new(
- 'collation' => $c, 'graph' => $safegraph );
# Now convert the witness groupings to a safe representation.
my $safe_groupings = [];
# 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( { 'linesep' => ' ' } ),
+ return { 'graph' => Text::Tradition::Stemma::editable_graph(
+ $safegraph, { 'linesep' => ' ' } ),
'groupings' => $safe_groupings };
}
}
sub _desanitize_names {
- my( $jsonstruct, $witness_map ) = @_;
+ my( $element, $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 );
- }
+ if( ref( $element ) eq 'ARRAY' ) {
+ foreach my $n ( @$element ) {
+ push( @$result, _desanitize_names( $n, $witness_map ) );
}
- push( @$result, $real_grouping );
+ } elsif( ref( $element ) eq 'HASH' ) {
+ my $real_hash = {};
+ map { $real_hash->{$_} = _desanitize_names( $element->{$_}, $witness_map ) }
+ keys %$element;
+ $result = $real_hash;
+ } elsif( exists $witness_map->{$element} ) {
+ $result = $witness_map->{$element}
+ } else {
+ $result = $element;
}
return $result;
}
+sub _invert_hash {
+ my( $hash ) = @_;
+ my $newhash;
+ foreach my $k ( keys %$hash ) {
+ if( ref( $hash->{$k} ) eq 'ARRAY' ) {
+ foreach my $v ( @{$hash->{$k}} ) {
+ $newhash->{$v} = $k;
+ }
+ } else {
+ $newhash->{$hash->{$k}} = $k;
+ }
+ }
+ return $newhash;
+}
+
### 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, $stemma, $variant_row, $lacunose ) = @_;
+ my $c = $tradition->collation;
# 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;
+
+ my $NO_IDP;
+ if( exists $variant_row->{'reading_roots'} ) {
+ $reading_roots = delete $variant_row->{'reading_roots'};
+ } else {
+ warn "No reading source information from IDP - proceed at your own risk";
+ $NO_IDP = 1;
+ }
+
+ # 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;
}
+ # Get the actual graph we should work with
+ my $graph;
+ try {
+ # contig contains all extant wits and all hypothetical wits
+ # needed to make up the groups.
+ $graph = $stemma->situation_graph( $contig, \@acwits );
+ } catch ( Text::Tradition::Error $e ) {
+ die "Could not extend graph with given extant and a.c. witnesses: "
+ . $e->message;
+ } catch {
+ die "Could not extend graph with a.c. witnesses @acwits";
+ }
+
+
# 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 @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 );
# Find the parent readings, if any, of this reading.
- my %rdgparents;
+ my $rdgparents = {};
foreach my $wit ( @roots ) {
- # Look in the main stemma to find this witness's extant or known-reading
+ # 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 olds.
my @check = $graph->predecessors( $wit );
while( @check ) {
my @next;
foreach my $wparent( @check ) {
my $preading = $contig->{$wparent};
- if( $preading ) {
- $rdgparents{$preading} = 1;
+ if( $preading && $preading ne $rid ) {
+ $rdgparents->{$preading} = 1;
} else {
push( @next, $graph->predecessors( $wparent ) );
}
@check = @next;
}
}
- $rdghash->{'reading_parents'} = [ keys %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, $rdghash->{readingid} );
+ 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 {
+ $phash->{relation} = { type => 'deletion' };
+ }
+ # 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;
+ }
+
+ $rdghash->{'reading_parents'} = $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;
+ my @trueroots;
+ if( exists $variant_row->{'classes'} ) {
+ # We have tested for reversions. Use the information.
+ my @reversions;
+ foreach my $rdgroot ( @roots ) {
+ ## TODO This needs IDP to prune itself in order to be
+ ## correct.
+ if( $variant_row->{'classes'}->{$rdgroot} eq 'revert' ) {
+ push( @reversions, $rdgroot );
+ } else {
+ push( @trueroots, $rdgroot );
+ }
+ }
+ $rdghash->{'independent_occurrence'} = \@trueroots;
+ $rdghash->{'reversion'} = \@reversions if @reversions;
+ } else {
+ @trueroots = @roots;
+ }
+ $rdghash->{'conflict'} = @trueroots != 1;
}
}
}
+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;
+}
-=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:
+=head2 similar( $word1, $word2 )
- [ [ 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.
-
+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
-sub perl_solver {
- my( $graph, @groups ) = @_;
- my @answer;
- foreach my $g ( @groups ) {
- push( @answer, _solve_variant_location( $graph, $g ) );
- }
- return \@answer;
-}
-
-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.
- }
+#!/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 {
- # 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.
- } else {
- push( @resolve, $wit );
- }
- } else {
- my $gst = pop @{$contig->{$wit}};
- $contig->{$wit} = $gst || '';
+ # Split the difference for substitutions
+ my $c1 = $diff->Range(1) || 1;
+ my $c2 = $diff->Range(2) || 1;
+ my $cd = ( $c1 + $c2 ) / 2;
+ $mag += $cd;
}
}
-
- 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;
+ return ( $mag <= length( $word1 ) / 2 );
}
sub _prune_group {
- my( $group, $stemma, $graph ) = @_;
- # 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;
+ my( $group, $graph ) = @_;
+ my $relevant = {};
+ # Record the existence of the vertices in the group
+ map { $relevant->{$_} = 1 } @$group;
# Make our subgraph
- my $subgraph = $graph->copy;
- map { $subgraph->delete_vertex( $_ ) unless exists $hypohash->{$_} }
+ my $subgraph = $graph->deep_copy;
+ map { $subgraph->delete_vertex( $_ ) unless $relevant->{$_} }
$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;
+ _prune_subtree( $subgraph );
+ # Return the list of vertices and the list of roots.
+ my $pruned_group = [ sort $subgraph->vertices ];
+ my $pruned_roots = [ $subgraph->predecessorless_vertices ];
+ return( $pruned_group, $pruned_roots );
}
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;
+ my( $tree ) = @_;
+
+ # Delete lacunose witnesses that have no successors
+ my @orphan_hypotheticals;
+ my $ctr = 0;
+ do {
+ die "Infinite loop on leaves" if $ctr > 100;
+ @orphan_hypotheticals =
+ grep { $tree->get_vertex_attribute( $_, 'class' ) eq 'hypothetical' }
+ $tree->successorless_vertices;
+ $tree->delete_vertices( @orphan_hypotheticals );
+ $ctr++;
+ } while( @orphan_hypotheticals );
+
+ # Delete lacunose roots that have a single successor
+ my @redundant_root;
+ $ctr = 0;
+ do {
+ die "Infinite loop on roots" if $ctr > 100;
+ @redundant_root =
+ grep { $tree->get_vertex_attribute( $_, 'class' ) eq 'hypothetical'
+ && $tree->successors( $_ ) == 1 }
+ $tree->predecessorless_vertices;
+ $tree->delete_vertices( @redundant_root );
+ $ctr++;
+ } while( @redundant_root );
}
sub _useful_variant {
projects / scpubgit/stemmatology.git / blobdiff