use Benchmark;
use Encode qw/ encode_utf8 /;
use Exporter 'import';
+use Graph;
use JSON qw/ encode_json decode_json /;
use LWP::UserAgent;
use Text::Tradition;
=item * merge_types - Specify a list of relationship types, where related readings
should be treated as identical for the purposes of analysis.
+=item * exclude_type1 - Exclude those ranks whose groupings have only type-1 variants.
+
=back
=begin testing
);
my $data = run_analysis( $tradition );
+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->{'conflict_count'}, 34, "Got right conflict count" );
is( $data->{'variant_count'}, 58, "Got right total variant number" );
+# TODO Make something meaningful of conflict count, maybe test other bits
=end testing
# Get the stemma
my $stemma = $tradition->stemma( $stemma_id );
- # Figure out which witnesses we are working with
+
+ # 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 = $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
# Group the variants to send to the solver
my @groups;
+ my @use_ranks;
my %lacunae;
foreach my $rank ( @ranks ) {
my $missing = [ @lacunose ];
- push( @groups, group_variants( $tradition, $rank, $missing, \@collapse ) );
+ my $rankgroup = group_variants( $tradition, $rank, $missing, \@collapse );
+ 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;
+ }
+ push( @use_ranks, $rank );
+ push( @groups, $rankgroup );
$lacunae{$rank} = $missing;
}
-
- # Parse the answer
+ # Run the solver
my $answer = solve_variants( $stemma, @groups );
# Do further analysis on the answer
my $conflict_count = 0;
- foreach my $idx ( 0 .. $#ranks ) {
+ my $aclabel = $c->ac_label;
+ foreach my $idx ( 0 .. $#use_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]};
+ $location->{'id'} = $use_ranks[$idx];
+ # 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 );
- # Add the reading text back in
+
+ # 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
my $rdg = $c->reading( $rdghash->{'readingid'} );
$rdghash->{'text'} = $rdg ? $rdg->text : $rdghash->{'readingid'};
+ # 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
}
}
$answer->{'conflict_count'} = $conflict_count;
be a reference to an array, to which the sigla of lacunose witnesses at this
rank will be appended.
-Returns two ordered lists $readings, $groups, where $readings->[$n] is attested
-by the witnesses listed in $groups->[$n].
+Returns a hash $group_readings where $rdg is attested by the witnesses listed
+in $group_readings->{$rdg}.
=cut
my( $tradition, $rank, $lacunose, $collapse ) = @_;
my $c = $tradition->collation;
my $aclabel = $c->ac_label;
+
# 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 $rdg = $tablewit->{'tokens'}->[$rank-1];
my $wit = $tablewit->{'witness'};
- $wit =~ s/^(.*)\Q$aclabel\E$/${1}_ac/;
+ # Exclude the witness if it is "lacunose" which if we got here
+ # means "not in the stemma".
+ next if $is_lacunose{$wit};
if( $rdg && $rdg->{'t'}->is_lacuna ) {
- _add_to_witlist( $wit, $lacunose, '_ac' );
+ _add_to_witlist( $wit, $lacunose, $aclabel );
} elsif( $rdg ) {
$readings_at_rank{$rdg->{'t'}->text} = $rdg->{'t'};
} else {
- _add_to_witlist( $wit, \@gap_wits, '_ac' );
+ _add_to_witlist( $wit, \@gap_wits, $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 ) {
+ 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};
+ # Get the witness list, including from readings collapsed into this one.
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 ) ) {
my @otherwits = $other->witnesses;
- map { s/\Q$aclabel\E$/_ac/ } @otherwits;
push( @wits, @otherwits );
$grouped_readings{$other->id} = 0;
}
}
- $grouped_readings{$rdg->id} = \@wits;
+ # 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{'(omitted)'} = \@gap_wits if @gap_wits;
# Get rid of our collapsed readings
keys %grouped_readings
if $collapse;
+ # Return the result
return \%grouped_readings;
}
+# 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 );
+ }
+ }
+}
+
=head2 solve_variants( $graph, @groups )
Sends the set of groups to the external graph solver service and returns
sub solve_variants {
my( $stemma, @groups ) = @_;
-
- # Make the json with stemma + groups
- my $jsonstruct = { 'graph' => $stemma->editable( ' ' ), 'groupings' => [] };
- foreach my $ghash ( @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
+
+ foreach my $idx ( 0..$#groups ) {
+ my $ghash = $groups[$idx];
my @grouping;
- foreach my $k ( sort keys %$ghash ) {
- push( @grouping, $ghash->{$k} );
+ # 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$/ } @sg );
+ push( @grouping, \@sg );
}
- push( @{$jsonstruct->{'groupings'}}, \@grouping );
+ # 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 );
}
- my $json = encode_json( $jsonstruct );
-
- # Send it off and get the result
+
+ ## 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();
- 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( $stemma, @{$jsonstruct->{'groupings'}} );
- }
-
- # Fold the result back into what we know about the groups.
- my $variants = [];
+ ## Witness map is a HACK to get around limitations in node names from IDP
+ my $witness_map = {};
+ ## Variables to store answers as they come back
+ my $variants = [ ( undef ) x ( scalar keys %$index_groupkeys ) ];
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;
+ 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 ) );
+ # Send it off and get the result
+ 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 );
}
- my $vstruct = {
- 'genealogical' => $result,
- 'readings' => [],
- };
- foreach my $k ( keys %$input_group ) {
- push( @{$vstruct->{'readings'}},
- { 'readingid' => $k, 'group' => $input_group->{$k}} );
+ ## 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;
+ }
+ }
+ # Retrieve the key for the original group that went to the solver
+ my $input_group = wit_stringify( $groupings->[$gidx] );
+ foreach my $oidx ( @{$group_indices->{$input_group}} ) {
+ my @readings = @{$index_groupkeys->{$oidx}};
+ my $vstruct = {
+ 'genealogical' => $result,
+ 'readings' => [],
+ };
+ foreach my $ridx ( 0 .. $#readings ) {
+ push( @{$vstruct->{'readings'}},
+ { 'readingid' => $readings[$ridx],
+ 'group' => $calc_groups->[$ridx] } );
+ }
+ $variants->[$oidx] = $vstruct;
+ }
}
- push( @$variants, $vstruct );
}
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( $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.
+ foreach my $n ( $graph->vertices ) {
+ my $sn = _safe_witstr( $n );
+ if( exists $witness_map->{$sn} ) {
+ warn "Ambiguous stringification $sn for $n and " . $witness_map->{$sn}
+ if $witness_map->{$sn} ne $n;
+ } else {
+ $witness_map->{$sn} = $n;
+ }
+ $safegraph->add_vertex( $sn );
+ $safegraph->set_vertex_attributes( $sn,
+ $graph->get_vertex_attributes( $n ) );
+ }
+ foreach my $e ( $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' => $c, '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( { 'linesep' => ' ' } ),
+ '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 );
+ }
+ return $result;
+}
+
+### END HACKERY ###
+
=head2 analyze_location ( $tradition, $graph, $location_hash )
Given the tradition, its stemma graph, and the solution from the graph solver,
=cut
sub perl_solver {
- my( $stemma, @groups ) = @_;
- my $graph = $stemma->graph;
+ my( $graph, @groups ) = @_;
my @answer;
foreach my $g ( @groups ) {
push( @answer, _solve_variant_location( $graph, $g ) );
return $variant_row;
}
+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;
+ # Make our subgraph
+ my $subgraph = $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 _prune_subtree {
my( $tree, $root, $contighash ) = @_;
# First, delete hypothetical leaves / orphans until there are none left.
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 );
- }
- }
-}
-
sub _symmdiff {
my( $lista, $listb ) = @_;
my %union;
projects / scpubgit/stemmatology.git / blobdiff