1 package Text::Tradition::Analysis;
6 use Text::Tradition::Stemma;
9 my( $class, $args ) = @_;
11 bless( $self, $class );
13 foreach my $t ( @{$args->{'traditions'}} ) {
14 $self->run_analysis( $t->{'file'}, $t->{'stemmadot'} );
20 my( $self, $file, $stemmadot ) = @_;
26 # Read in the file and stemma
27 my $tradition = Text::Tradition->new(
32 $data->{'title'} = $tradition->name;
34 my $stemma = Text::Tradition::Stemma->new(
35 'collation' => $tradition->collation,
38 # We will return the stemma picture
39 $svg = $stemma->as_svg( { size => "8,7.5" } );;
40 $data->{'svg'} = $svg;
42 # We have the collation, so get the alignment table with witnesses in rows.
43 # Also return the reading objects in the table, rather than just the words.
45 map { $wits->{$_} = 1 } $stemma->witnesses;
46 my $all_wits_table = $tradition->collation->make_alignment_table( 'refs', $wits );
48 # For each column in the alignment table, we want to see if the existing
49 # groupings of witnesses match our stemma hypothesis. We also want, at the
50 # end, to produce an HTML table with all the variants.
52 my ( $total, $genealogical, $conflicts ) = ( 0, 0, 0 );
54 # Strip the list of sigla and save it for correlation to the readings.
55 my $col_wits = shift @$all_wits_table;
56 # Any witness in the stemma that has no row should be noted.
57 foreach ( @$col_wits ) {
58 $wits->{$_}++; # Witnesses present in table and stemma now have value 2.
60 my @not_collated = grep { $wits->{$_} == 1 } keys %$wits;
62 # We will return a data structure, an array for each row that looks like:
63 # { id = X, genealogical = Y, readings = [ text = X, group = Y], empty = N }
64 foreach my $i ( 0 .. $#$all_wits_table ) {
65 # For each column in the table, group the readings by witness.
67 my $col_rdgs = shift @$all_wits_table;
69 my $lacunose = [ @not_collated ];
70 foreach my $j ( 0 .. $#{$col_rdgs} ) {
71 my $rdg = $col_rdgs->[$j];
72 my $rdg_text = '(omitted)'; # Initialize in case of empty reading
74 if( $rdg->is_lacuna ) {
75 $rdg_text = undef; # Don't count lacunae
76 push( @$lacunose, $col_wits->[$j] );
78 $rdg_text = $rdg->text;
79 # Get the rank from any real reading; they should be identical.
83 if( defined $rdg_text ) {
84 # Initialize the witness array if we haven't got one yet
85 $rdg_wits->{$rdg_text} = [] unless $rdg_wits->{$rdg_text};
86 # Add the relevant witness, subject to a.c. logic
87 add_variant_wit( $rdg_wits->{$rdg_text}, $col_wits->[$j],
88 $tradition->collation->ac_label );
92 # See if this column has any potentially genealogical variants.
93 # If not, skip to the next.
94 $total++ unless scalar keys %$rdg_wits == 1;
95 my( $groups, $readings ) = useful_variant( $rdg_wits );
96 next unless $groups && $readings;
98 # Keep track of our widest row
99 $html_columns = scalar @$groups if scalar @$groups > $html_columns;
101 # We can already look up witnesses for a reading; we also want to look
102 # up readings for a given witness.
103 my $group_readings = {};
104 foreach my $x ( 0 .. $#$groups ) {
105 $group_readings->{wit_stringify( $groups->[$x] )} = $readings->[$x];
108 # For all the groups with more than one member, collect the list of all
109 # contiguous vertices needed to connect them.
111 my $variant_row = analyze_variant_location( $group_readings, $groups,
112 $stemma->graph, $lacunose );
113 $variant_row->{'id'} = $rank;
114 $genealogical++ if $variant_row->{'genealogical'};
115 $conflicts += grep { $_->{'conflict'} } @{$variant_row->{'readings'}};
117 # Now run the same analysis given the calculated distance tree(s).
118 # my @trees = @{$stemma->distance_trees};
120 # foreach my $tree ( 0 .. $#trees ) {
121 # my $dc = analyze_variant_location( $group_readings, $groups, $tree, $lacunose, 'undirected' );
122 # foreach my $rdg ( keys %$dc ) {
123 # my $var = $dc->{$rdg};
124 # # TODO Do something with this
129 # Record that we used this variant in an analysis
130 push( @$variants, $variant_row );
133 # Go through our variant rows, after we have seen all of them once,
134 # and add the number of empty columns needed by each.
135 foreach my $row ( @$variants ) {
136 my $empty = $html_columns - scalar @{$row->{'readings'}};
137 $row->{'empty'} = $empty;
140 # Populate self with our analysis data.
141 $data->{'variants'} = $variants;
142 $data->{'variant_count'} = $total;
143 $data->{'conflict_count'} = $conflicts;
144 $data->{'genealogical_count'} = $genealogical;
145 push( @{$self->{'data'}}, $data );
148 # variant_row -> genealogical
149 # -> readings [ { text, group, conflict, missing } ]
151 sub analyze_variant_location {
152 my( $group_readings, $groups, $graph, $lacunose, $undirected ) = @_;
158 map { $missing->{$_} = 1 } @$lacunose;
159 my $variant_row = { 'readings' => [] };
160 # Mark each ms as in its own group, first.
161 foreach my $g ( @$groups ) {
162 my $gst = wit_stringify( $g );
163 map { $contig->{$_} = $gst } @$g;
165 # Now for each unmarked node in the graph, initialize an array
166 # for possible group memberships. We will use this later to
167 # resolve potential conflicts.
168 map { $contig->{$_} = [] unless $contig->{$_} } $graph->vertices;
169 foreach my $g ( sort { scalar @$b <=> scalar @$a } @$groups ) {
170 my $gst = wit_stringify( $g ); # This is the group name
171 my $reachable = { $g->[0] => 1 };
172 # Copy the graph, and delete all non-members from the new graph.
173 my $part = $graph->copy;
175 $part->delete_vertices(
176 grep { !ref( $contig->{$_} ) && $contig->{$_} ne $gst } $graph->vertices );
178 # Now look to see if our group is connected.
179 if( $undirected ) { # For use with distance trees etc.
180 # Find all vertices reachable from the first (arbitrary) group
181 # member. If we are genealogical this should include them all.
182 map { $reachable->{$_} = 1 } $part->all_reachable( $g->[0] );
183 # TODO This is a terrible way to do distance trees, since all
184 # non-leaf nodes are included in every graph part now. We may
185 # have to go back to SPDP.
188 # Dispense with the trivial case of one reading.
189 # We have to take directionality into account.
190 # How many root nodes do we have?
191 my @roots = grep { ref( $contig->{$_} ) || $contig->{$_} eq $gst }
192 $part->source_vertices;
193 # Assuming that @$g > 1, find the first root node that has at
194 # least one successor belonging to our group. If this reading
195 # is genealogical, there should be only one, but we will check
196 # that implicitly later.
197 my $nodes_in_subtree = 0;
198 foreach my $root ( @roots ) {
199 # Prune the tree to get rid of extraneous hypotheticals.
200 $root = prune_subtree( $part, $root, $contig );
201 # Get all the successor nodes of our root.
202 my $tmp_reach = { $root => 1 };
203 map { $tmp_reach->{$_} = 1 } $part->all_successors( $root );
204 # Skip this root if none of our successors are in our group
205 # (e.g. isolated 'hypothetical' witnesses with no group)
206 next unless grep { $contig->{$_} } keys %$tmp_reach;
207 if( keys %$tmp_reach > $nodes_in_subtree ) {
208 $nodes_in_subtree = keys %$tmp_reach;
209 $reachable = $tmp_reach;
213 } # else it is a single-node group, nothing to calculate.
216 # None of the 'reachable' nodes should be marked as being in another
217 # group. Paint the 'hypotheticals' with our group while we are at it,
218 # unless there is a conflict present.
219 foreach ( keys %$reachable ) {
220 if( ref $contig->{$_} ) {
221 push( @{$contig->{$_}}, $gst );
222 } elsif( $contig->{$_} ne $gst ) {
223 $conflict->{$group_readings->{$gst}} = $group_readings->{$contig->{$_}};
224 } # else it is an 'extant' node marked with our group already.
226 # None of the unreachable nodes should be in our group either.
227 foreach ( $part->vertices ) {
228 next if $reachable->{$_};
229 if( $contig->{$_} eq $gst ) {
230 $conflict->{$group_readings->{$gst}} = $group_readings->{$gst};
235 # Now, if we have a conflict, we can write the reading in full. If not,
236 # we have to save the subgraph so that we can resolve possible conflicts
237 # on hypothetical nodes.
238 $is_conflicted = 1 if exists $conflict->{$group_readings->{$gst}};
241 my $reading = { 'text' => $group_readings->{$gst},
242 'missing' => wit_stringify( $lacunose ),
243 'group' => $gst }; # This will change if we find no conflict
244 if( $is_conflicted ) {
245 $reading->{'conflict'} = $conflict->{$group_readings->{$gst}}
247 # Save the relevant subgraph.
248 $subgraph->{$gst} = { 'graph' => $part,
249 'root' => $group_root,
250 'reachable' => $reachable };
252 push( @{$variant_row->{'readings'}}, $reading );
255 # Now that we have gone through all the rows, check the hypothetical
256 # readings for conflict if we haven't found one yet.
257 if( keys %$subgraph && !keys %$conflict ) {
259 foreach ( keys %$contig ) {
260 next unless ref $contig->{$_};
261 if( scalar @{$contig->{$_}} > 1 ) {
262 push( @resolve, $_ );
264 $contig->{$_} = scalar @{$contig->{$_}} ? $contig->{$_}->[0] : '';
267 # Do we still have a possible conflict?
268 my $still_contig = {};
269 foreach my $h ( @resolve ) {
270 # For each of the hypothetical readings with more than one possibility,
271 # try deleting it from each of its member subgraphs in turn, and see
272 # if that breaks the contiguous grouping.
273 # TODO This can still break in a corner case where group A can use
274 # either vertex 1 or 2, and group B can use either vertex 2 or 1.
275 # Revisit this if necessary; it could get brute-force nasty.
276 foreach my $gst ( @{$contig->{$h}} ) {
277 my $gpart = $subgraph->{$gst}->{'graph'}->copy;
278 my $reachable = $subgraph->{$gst}->{'reachable'};
279 $gpart->delete_vertex( $h );
280 # Is everything else still reachable from the root?
281 # TODO If $h was the root, see if we still have a single root.
282 my %still_reachable = ( $subgraph->{$gst}->{'root'} => 1 );
283 map { $still_reachable{$_} = 1 }
284 $gpart->all_successors( $subgraph->{$gst}->{'root'} );
285 foreach my $v ( keys %$reachable ) {
287 if( !$still_reachable{$v}
288 && ( $contig->{$v} eq $gst
289 || ( exists $still_contig->{$v}
290 && $still_contig->{$v} eq $gst ) ) ) {
292 if( exists $still_contig->{$h} ) {
294 $conflict->{$group_readings->{$gst}} =
295 $group_readings->{$still_contig->{$h}};
297 $still_contig->{$h} = $gst;
300 } # else we don't need $h in this group.
305 # Now, assuming no conflict, we have some hypothetical vertices in
306 # $still_contig that are the "real" group memberships. Replace these
308 unless ( keys %$conflict ) {
309 foreach my $v ( keys %$contig ) {
310 next unless ref $contig->{$v};
311 $contig->{$v} = $still_contig->{$v};
316 # Now write the group and conflict information into the respective rows.
317 foreach my $rdg ( @{$variant_row->{'readings'}} ) {
318 $rdg->{'conflict'} = $conflict->{$rdg->{'text'}};
319 next if $rdg->{'conflict'};
320 my @members = grep { $contig->{$_} eq $rdg->{'group'} && !$missing->{$_} }
322 $rdg->{'group'} = wit_stringify( \@members );
325 $variant_row->{'genealogical'} = !( keys %$conflict );
330 my( $tree, $root, $contighash ) = @_;
331 # First, delete hypothetical leaves / orphans until there are none left.
332 my @orphan_hypotheticals = grep { ref( $contighash->{$_} ) }
333 $tree->successorless_vertices;
334 while( @orphan_hypotheticals ) {
335 $tree->delete_vertices( @orphan_hypotheticals );
336 @orphan_hypotheticals = grep { ref( $contighash->{$_} ) }
337 $tree->successorless_vertices;
339 # Then delete a hypothetical root with only one successor, moving the
341 while( $tree->successors( $root ) == 1 && ref $contighash->{$root} ) {
342 my @nextroot = $tree->successors( $root );
343 $tree->delete_vertex( $root );
344 $root = $nextroot[0];
346 # The tree has been modified in place, but we need to know the new root.
349 # Add the variant, subject to a.c. representation logic.
350 # This assumes that we will see the 'main' version before the a.c. version.
351 sub add_variant_wit {
352 my( $arr, $wit, $acstr ) = @_;
354 if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
356 $skip = grep { $_ =~ /^\Q$real\E$/ } @$arr;
358 push( @$arr, $wit ) unless $skip;
361 # Return an answer if the variant is useful, i.e. if there are at least 2 variants
362 # with at least 2 witnesses each.
364 my( $readings ) = @_;
365 my $total = keys %$readings;
366 foreach my $var ( keys %$readings ) {
367 $total-- if @{$readings->{$var}} == 1;
369 return( undef, undef ) if $total <= 1;
370 my( $groups, $text );
371 foreach my $var ( keys %$readings ) {
372 push( @$groups, $readings->{$var} );
373 push( @$text, $var );
375 return( $groups, $text );
378 # Take an array of witness groupings and produce a string like
379 # ['A','B'] / ['C','D','E'] / ['F']
384 # If we were passed an array of witnesses instead of an array of
385 # groupings, then "group" the witnesses first.
386 unless( ref( $groups->[0] ) ) {
387 my $mkgrp = [ $groups ];
390 foreach my $g ( @$groups ) {
391 push( @gst, '[' . join( ',', map { "'$_'" } @$g ) . ']' );
393 return join( ' / ', @gst );