1 package Text::Tradition::Analysis;
8 use Text::Tradition::Stemma;
10 use vars qw/ @EXPORT_OK /;
11 @EXPORT_OK = qw/ run_analysis group_variants analyze_variant_location wit_stringify /;
14 my( $tradition ) = @_;
19 # We need a stemma in order to run this...
20 unless( $tradition->stemma_count ) {
21 warn "Tradition '" . $tradition->name . "' has no stemma to analyze";
24 my $stemma = $tradition->stemma(0); # TODO allow multiple
26 # We have the collation, so get the alignment table with witnesses in rows.
27 # Also return the reading objects in the table, rather than just the words.
29 map { $wits->{$_} = 1 } $stemma->witnesses;
30 # For each column in the alignment table, we want to see if the existing
31 # groupings of witnesses match our stemma hypothesis. We also need to keep
32 # track of the maximum number of variants at any one location.
34 my ( $genealogical, $conflicts ) = ( 0, 0, 0 );
36 my $variant_groups = group_variants( $tradition->collation, $wits );
37 foreach my $rank ( 0 .. $#{$variant_groups} ) {
38 my $groups = $variant_groups->[$rank]->{'groups'};
39 my $readings = $variant_groups->[$rank]->{'readings'};
40 my $lacunose = $variant_groups->[$rank]->{'lacunose'};
42 $max_variants = scalar @$groups if scalar @$groups > $max_variants;
44 # We can already look up witnesses for a reading; we also want to look
45 # up readings for a given witness.
46 my $group_readings = {};
47 foreach my $x ( 0 .. $#$groups ) {
48 $group_readings->{wit_stringify( $groups->[$x] )} = $readings->[$x];
51 # For all the groups with more than one member, collect the list of all
52 # contiguous vertices needed to connect them.
53 my $variant_loc = analyze_variant_location( $group_readings, $groups,
54 $stemma->graph, $lacunose );
55 $variant_loc->{'id'} = $rank;
56 $genealogical++ if $variant_loc->{'genealogical'};
57 $conflicts += grep { $_->{'conflict'} } @{$variant_loc->{'readings'}};
59 # Now run the same analysis given the calculated distance tree(s).
60 # my @trees = @{$stemma->distance_trees};
62 # foreach my $tree ( 0 .. $#trees ) {
63 # my $dc = analyze_variant_location( $group_readings, $groups, $tree, $lacunose, 'undirected' );
64 # foreach my $rdg ( keys %$dc ) {
65 # my $var = $dc->{$rdg};
66 # # TODO Do something with this
71 # Record that we used this variant in an analysis
72 push( @$variants, $variant_loc );
75 # Go through our variant locations, after we have seen all of them once,
76 # and add the number of empty columns needed by each.
77 foreach my $row ( @$variants ) {
78 my $empty = $max_variants - scalar @{$row->{'readings'}};
79 $row->{'empty'} = $empty;
82 $data->{'variants'} = $variants;
83 $data->{'variant_count'} = $tradition->collation->end->rank - 1;
84 $data->{'conflict_count'} = $conflicts;
85 $data->{'genealogical_count'} = $genealogical;
91 my $variant_groups = [];
93 # We have the collation, so get the alignment table with witnesses in rows.
94 # Also return the reading objects in the table, rather than just the words.
95 my $all_wits_table = $c->make_alignment_table( 'refs', $wits );
96 # Strip the list of sigla and save it for correlation to the readings.
97 my @table_wits = map { $_->{'witness'} } @{$all_wits_table->{'alignment'}};
98 # Any witness in the stemma that has no row should be noted.
99 foreach ( @table_wits ) {
100 $wits->{$_}++; # Witnesses present in table and stemma now have value 2.
102 my @not_collated = grep { $wits->{$_} == 1 } keys %$wits;
103 foreach my $i ( 0 .. $all_wits_table->{'length'} - 1 ) {
104 # For each column in the table, group the readings by witness.
106 my @col_rdgs = map { $_->{tokens}->[$i] } @{$all_wits_table->{'alignment'}};
107 my $lacunose = [ @not_collated ];
108 foreach my $j ( 0 .. $#col_rdgs ) {
109 my $rdg = $col_rdgs[$j];
110 my $rdg_text = '(omitted)'; # Initialize in case of empty reading
112 if( $rdg->{'t'}->is_lacuna ) {
113 $rdg_text = undef; # Don't count lacunae
114 push( @$lacunose, $table_wits[$j] );
116 $rdg_text = $rdg->{'t'}->text;
119 if( defined $rdg_text ) {
120 # Initialize the witness array if we haven't got one yet
121 $rdg_wits->{$rdg_text} = [] unless $rdg_wits->{$rdg_text};
122 # Add the relevant witness, subject to a.c. logic
123 add_variant_wit( $rdg_wits->{$rdg_text}, $table_wits[$j],
128 # See if this column has any potentially genealogical variants.
129 # If not, skip to the next.
130 my( $groups, $readings ) = useful_variant( $rdg_wits );
131 next unless $groups && $readings;
133 push( @$variant_groups,
134 { 'groups' => $groups, 'readings' => $readings, 'lacunose' => $lacunose } );
136 return $variant_groups;
141 # variant_row -> genealogical
142 # -> readings [ { text, group, conflict, missing } ]
144 sub analyze_variant_location {
145 my( $group_readings, $groups, $graph, $lacunose, $undirected ) = @_;
151 map { $missing->{$_} = 1 } @$lacunose;
152 my $variant_row = { 'readings' => [] };
153 # Mark each ms as in its own group, first.
154 foreach my $g ( @$groups ) {
155 my $gst = wit_stringify( $g );
156 map { $contig->{$_} = $gst } @$g;
158 # Now for each unmarked node in the graph, initialize an array
159 # for possible group memberships. We will use this later to
160 # resolve potential conflicts.
161 map { $contig->{$_} = [] unless $contig->{$_} } $graph->vertices;
162 foreach my $g ( sort { scalar @$b <=> scalar @$a } @$groups ) {
163 my $gst = wit_stringify( $g ); # This is the group name
164 my $reachable = { $g->[0] => 1 };
165 # Copy the graph, and delete all non-members from the new graph.
166 my $part = $graph->copy;
168 $part->delete_vertices(
169 grep { !ref( $contig->{$_} ) && $contig->{$_} ne $gst } $graph->vertices );
171 # Now look to see if our group is connected.
172 if( $undirected ) { # For use with distance trees etc.
173 # Find all vertices reachable from the first (arbitrary) group
174 # member. If we are genealogical this should include them all.
175 map { $reachable->{$_} = 1 } $part->all_reachable( $g->[0] );
176 # TODO This is a terrible way to do distance trees, since all
177 # non-leaf nodes are included in every graph part now. We may
178 # have to go back to SPDP.
181 # Dispense with the trivial case of one reading.
182 # We have to take directionality into account.
183 # How many root nodes do we have?
184 my @roots = grep { ref( $contig->{$_} ) || $contig->{$_} eq $gst }
185 $part->source_vertices;
186 # Assuming that @$g > 1, find the first root node that has at
187 # least one successor belonging to our group. If this reading
188 # is genealogical, there should be only one, but we will check
189 # that implicitly later.
190 my $nodes_in_subtree = 0;
191 foreach my $root ( @roots ) {
192 # Prune the tree to get rid of extraneous hypotheticals.
193 $root = prune_subtree( $part, $root, $contig );
194 # Get all the successor nodes of our root.
195 my $tmp_reach = { $root => 1 };
196 map { $tmp_reach->{$_} = 1 } $part->all_successors( $root );
197 # Skip this root if none of our successors are in our group
198 # (e.g. isolated 'hypothetical' witnesses with no group)
199 next unless grep { $contig->{$_} } keys %$tmp_reach;
200 if( keys %$tmp_reach > $nodes_in_subtree ) {
201 $nodes_in_subtree = keys %$tmp_reach;
202 $reachable = $tmp_reach;
206 } # else it is a single-node group, nothing to calculate.
209 # None of the 'reachable' nodes should be marked as being in another
210 # group. Paint the 'hypotheticals' with our group while we are at it,
211 # unless there is a conflict present.
212 foreach ( keys %$reachable ) {
213 if( ref $contig->{$_} ) {
214 push( @{$contig->{$_}}, $gst );
215 } elsif( $contig->{$_} ne $gst ) {
216 $conflict->{$group_readings->{$gst}} = $group_readings->{$contig->{$_}};
217 } # else it is an 'extant' node marked with our group already.
219 # None of the unreachable nodes should be in our group either.
220 foreach ( $part->vertices ) {
221 next if $reachable->{$_};
222 if( $contig->{$_} eq $gst ) {
223 $conflict->{$group_readings->{$gst}} = $group_readings->{$gst};
228 # Now, if we have a conflict, we can write the reading in full. If not,
229 # we have to save the subgraph so that we can resolve possible conflicts
230 # on hypothetical nodes.
231 $is_conflicted = 1 if exists $conflict->{$group_readings->{$gst}};
234 my $reading = { 'text' => $group_readings->{$gst},
235 'missing' => wit_stringify( $lacunose ),
236 'group' => $gst }; # This will change if we find no conflict
237 if( $is_conflicted ) {
238 $reading->{'conflict'} = $conflict->{$group_readings->{$gst}}
240 # Save the relevant subgraph.
241 $subgraph->{$gst} = { 'graph' => $part,
242 'root' => $group_root,
243 'reachable' => $reachable };
245 push( @{$variant_row->{'readings'}}, $reading );
248 # Now that we have gone through all the rows, check the hypothetical
249 # readings for conflict if we haven't found one yet.
250 if( keys %$subgraph && !keys %$conflict ) {
252 foreach ( keys %$contig ) {
253 next unless ref $contig->{$_};
254 if( scalar @{$contig->{$_}} > 1 ) {
255 push( @resolve, $_ );
257 $contig->{$_} = scalar @{$contig->{$_}} ? $contig->{$_}->[0] : '';
260 # Do we still have a possible conflict?
261 my $still_contig = {};
262 foreach my $h ( @resolve ) {
263 # For each of the hypothetical readings with more than one possibility,
264 # try deleting it from each of its member subgraphs in turn, and see
265 # if that breaks the contiguous grouping.
266 # TODO This can still break in a corner case where group A can use
267 # either vertex 1 or 2, and group B can use either vertex 2 or 1.
268 # Revisit this if necessary; it could get brute-force nasty.
269 foreach my $gst ( @{$contig->{$h}} ) {
270 my $gpart = $subgraph->{$gst}->{'graph'}->copy;
271 my $reachable = $subgraph->{$gst}->{'reachable'};
272 $gpart->delete_vertex( $h );
273 # Is everything else still reachable from the root?
274 # TODO If $h was the root, see if we still have a single root.
275 my %still_reachable = ( $subgraph->{$gst}->{'root'} => 1 );
276 map { $still_reachable{$_} = 1 }
277 $gpart->all_successors( $subgraph->{$gst}->{'root'} );
278 foreach my $v ( keys %$reachable ) {
280 if( !$still_reachable{$v}
281 && ( $contig->{$v} eq $gst
282 || ( exists $still_contig->{$v}
283 && $still_contig->{$v} eq $gst ) ) ) {
285 if( exists $still_contig->{$h} ) {
287 $conflict->{$group_readings->{$gst}} =
288 $group_readings->{$still_contig->{$h}};
290 $still_contig->{$h} = $gst;
293 } # else we don't need $h in this group.
298 # Now, assuming no conflict, we have some hypothetical vertices in
299 # $still_contig that are the "real" group memberships. Replace these
301 unless ( keys %$conflict ) {
302 foreach my $v ( keys %$contig ) {
303 next unless ref $contig->{$v};
304 $contig->{$v} = $still_contig->{$v};
309 # Now write the group and conflict information into the respective rows.
310 foreach my $rdg ( @{$variant_row->{'readings'}} ) {
311 $rdg->{'conflict'} = $conflict->{$rdg->{'text'}};
312 next if $rdg->{'conflict'};
313 my @members = grep { $contig->{$_} eq $rdg->{'group'} && !$missing->{$_} }
315 $rdg->{'group'} = wit_stringify( \@members );
318 $variant_row->{'genealogical'} = !( keys %$conflict );
323 my( $tree, $root, $contighash ) = @_;
324 # First, delete hypothetical leaves / orphans until there are none left.
325 my @orphan_hypotheticals = grep { ref( $contighash->{$_} ) }
326 $tree->successorless_vertices;
327 while( @orphan_hypotheticals ) {
328 $tree->delete_vertices( @orphan_hypotheticals );
329 @orphan_hypotheticals = grep { ref( $contighash->{$_} ) }
330 $tree->successorless_vertices;
332 # Then delete a hypothetical root with only one successor, moving the
334 while( $tree->successors( $root ) == 1 && ref $contighash->{$root} ) {
335 my @nextroot = $tree->successors( $root );
336 $tree->delete_vertex( $root );
337 $root = $nextroot[0];
339 # The tree has been modified in place, but we need to know the new root.
342 # Add the variant, subject to a.c. representation logic.
343 # This assumes that we will see the 'main' version before the a.c. version.
344 sub add_variant_wit {
345 my( $arr, $wit, $acstr ) = @_;
347 if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
349 $skip = grep { $_ =~ /^\Q$real\E$/ } @$arr;
351 push( @$arr, $wit ) unless $skip;
354 # Return an answer if the variant is useful, i.e. if there are at least 2 variants
355 # with at least 2 witnesses each.
357 my( $readings ) = @_;
358 my $total = keys %$readings;
359 foreach my $var ( keys %$readings ) {
360 $total-- if @{$readings->{$var}} == 1;
362 return( undef, undef ) if $total <= 1;
363 my( $groups, $text );
364 foreach my $var ( keys %$readings ) {
365 push( @$groups, $readings->{$var} );
366 push( @$text, $var );
368 return( $groups, $text );
371 # Take an array of witness groupings and produce a string like
372 # ['A','B'] / ['C','D','E'] / ['F']
377 # If we were passed an array of witnesses instead of an array of
378 # groupings, then "group" the witnesses first.
379 unless( ref( $groups->[0] ) ) {
380 my $mkgrp = [ $groups ];
383 foreach my $g ( @$groups ) {
384 push( @gst, '[' . join( ',', map { "'$_'" } @$g ) . ']' );
386 return join( ' / ', @gst );