X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FText%2FTradition%2FAnalysis.pm;h=fda51376077c9e67781e1c5e134c203704fd706a;hb=f2fec47dc44816775c844e8b50cef03b12b69dae;hp=7dd592a93ff69a133a1edee9f4af8f09b39401b7;hpb=e59b8faa0a5eb60c66c68cfabbe25aed31e8c850;p=scpubgit%2Fstemmatology.git diff --git a/lib/Text/Tradition/Analysis.pm b/lib/Text/Tradition/Analysis.pm index 7dd592a..fda5137 100644 --- a/lib/Text/Tradition/Analysis.pm +++ b/lib/Text/Tradition/Analysis.pm @@ -5,6 +5,7 @@ use warnings; 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; @@ -55,6 +56,8 @@ is 0 (i.e. the first). =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 @@ -109,8 +112,8 @@ foreach my $row ( @{$data->{'variants'}} ) { is( $row->{'genealogical'}, $expected_genealogical{$row->{'id'}}, "Got correct genealogical flag for row " . $row->{'id'} ); } -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 @@ -121,15 +124,16 @@ sub run_analysis { my $c = $tradition->collation; my $stemma_id = $opts{'stemma_id'} || 0; - my @ranks = @{$opts{'ranks'}} if ref( $opts{'ranks'} ) eq 'ARRAY'; - my @collapse = @{$opts{'merge_types'}} if ref( $opts{'merge_types'} ) eq 'ARRAY'; + my @ranks = ref( $opts{'ranks'} ) eq 'ARRAY' ? @{$opts{'ranks'}} : (); + my @collapse = ref( $opts{'merge_types'} ) eq 'ARRAY' ? @{$opts{'merge_types'}} : (); # Get the stemma my $stemma = $tradition->stemma( $stemma_id ); + # Figure out which witnesses we are working with my @lacunose = $stemma->hypotheticals; my @tradition_wits = map { $_->sigil } $tradition->witnesses; - map { push( @tradition_wits, $_->sigil."_ac" ) if $_->is_layered } + map { push( @tradition_wits, $_->sigil.$c->ac_label ) if $_->is_layered } $tradition->witnesses; push( @lacunose, _symmdiff( [ $stemma->witnesses ], \@tradition_wits ) ); @@ -148,10 +152,16 @@ sub run_analysis { 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( $rankgroup ); + next unless @$rdgs; + } + push( @groups, $rankgroup ); $lacunae{$rank} = $missing; } - + $DB::single = 1; # Parse the answer my $answer = solve_variants( $stemma, @groups ); @@ -185,8 +195,8 @@ relationships in @merge_relationship_types as equivalent. $lacunose should be a reference to an array, to which the sigla of lacunose witnesses at this rank will be appended. -Returns two ordered lists $readings, $groups, where $readings->[$n] is attested -by the witnesses listed in $groups->[$n]. +Returns a hash $group_readings where $rdg is attested by the witnesses listed +in $group_readings->{$rdg}. =cut @@ -194,16 +204,13 @@ by the witnesses listed in $groups->[$n]. sub group_variants { my( $tradition, $rank, $lacunose, $collapse ) = @_; my $c = $tradition->collation; - # All the regexps here are to get rid of space characters in witness names. - my $aclabel = $c->ac_label; - $aclabel =~ s/\s/_/g; + my $aclabel = $c->ac_label; # Get the alignment table readings my %readings_at_rank; my @gap_wits; - foreach my $tablewit ( @{$tradition->collation->alignment_table->{'alignment'}} ) { + foreach my $tablewit ( @{$c->alignment_table->{'alignment'}} ) { my $rdg = $tablewit->{'tokens'}->[$rank-1]; my $wit = $tablewit->{'witness'}; - $wit =~ s/\s/_/g; if( $rdg && $rdg->{'t'}->is_lacuna ) { _add_to_witlist( $wit, $lacunose, $aclabel ); } elsif( $rdg ) { @@ -215,16 +222,15 @@ sub group_variants { # 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 ( sort { $b->witnesses <=> $a->witnesses } + 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; - map { s/\s/_/g } @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/\s/_/g } @otherwits; push( @wits, @otherwits ); $grouped_readings{$other->id} = 0; } @@ -261,15 +267,17 @@ sub solve_variants { my( $stemma, @groups ) = @_; # Make the json with stemma + groups - my $jsonstruct = { 'graph' => $stemma->editable( ' ' ), 'groupings' => [] }; + my $groupings = []; foreach my $ghash ( @groups ) { my @grouping; - foreach my $k ( sort keys %$ghash ) { + foreach my $k ( keys %$ghash ) { push( @grouping, $ghash->{$k} ); } - push( @{$jsonstruct->{'groupings'}}, \@grouping ); + push( @$groupings, \@grouping ); } - my $json = encode_json( $jsonstruct ); + ## 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'; @@ -278,13 +286,15 @@ sub solve_variants { 'Content' => $json ); my $answer; + my $used_idp; if( $resp->is_success ) { - $answer = decode_json( $resp->content ); + $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( $stemma, @{$jsonstruct->{'groupings'}} ); + $answer = perl_solver( $stemma, @$groupings ); } # Fold the result back into what we know about the groups. @@ -292,7 +302,18 @@ sub solve_variants { my $genealogical = 0; foreach my $idx ( 0 .. $#groups ) { my( $calc_groups, $result ) = @{$answer->[$idx]}; - $genealogical++ if $result; + 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 $k ( sort keys %$input_group ) { my $cg = shift @$calc_groups; @@ -314,6 +335,88 @@ sub solve_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 ); + } + return $result; +} + +### END HACKERY ### + =head2 analyze_location ( $tradition, $graph, $location_hash ) Given the tradition, its stemma graph, and the solution from the graph solver, @@ -606,6 +709,24 @@ sub _solve_variant_location { 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 _prune_subtree { my( $tree, $root, $contighash ) = @_; # First, delete hypothetical leaves / orphans until there are none left.