1 package Text::Tradition::Analysis;
8 use Text::Tradition::Stemma;
10 use vars qw/ @EXPORT_OK /;
11 @EXPORT_OK = qw/ run_analysis group_variants wit_stringify /;
14 my( $class, $args ) = @_;
16 bless( $self, $class );
18 foreach my $t ( @{$args->{'traditions'}} ) {
19 $self->run_analysis( $t->{'file'}, $t->{'stemmadot'} );
25 my( $self, $file, $stemmadot ) = @_;
31 # Read in the file and stemma
32 my $tradition = Text::Tradition->new(
37 $data->{'title'} = $tradition->name;
39 my $stemma = Text::Tradition::Stemma->new(
40 'collation' => $tradition->collation,
43 # We will return the stemma picture
44 $svg = $stemma->as_svg( { size => "8,7.5" } );;
45 $data->{'svg'} = $svg;
47 # We have the collation, so get the alignment table with witnesses in rows.
48 # Also return the reading objects in the table, rather than just the words.
50 map { $wits->{$_} = 1 } $stemma->witnesses;
51 my $all_wits_table = $tradition->collation->make_alignment_table( 'refs', $wits );
53 # For each column in the alignment table, we want to see if the existing
54 # groupings of witnesses match our stemma hypothesis. We also want, at the
55 # end, to produce an HTML table with all the variants.
57 my ( $total, $genealogical, $conflicts ) = ( 0, 0, 0 );
59 # Strip the list of sigla and save it for correlation to the readings.
60 my $col_wits = shift @$all_wits_table;
61 # Any witness in the stemma that has no row should be noted.
62 foreach ( @$col_wits ) {
63 $wits->{$_}++; # Witnesses present in table and stemma now have value 2.
65 my @not_collated = grep { $wits->{$_} == 1 } keys %$wits;
67 # We will return a data structure, an array for each row that looks like:
68 # { id = X, genealogical = Y, readings = [ text = X, group = Y], empty = N }
70 my $t0 = Benchmark->new();
71 foreach my $i ( 0 .. $#$all_wits_table ) {
72 # For each column in the table, group the readings by witness.
74 my $col_rdgs = shift @$all_wits_table;
76 my $lacunose = [ @not_collated ];
77 foreach my $j ( 0 .. $#{$col_rdgs} ) {
78 my $rdg = $col_rdgs->[$j];
79 my $rdg_text = '(omitted)'; # Initialize in case of empty reading
81 if( $rdg->is_lacuna ) {
82 $rdg_text = undef; # Don't count lacunae
83 push( @$lacunose, $col_wits->[$j] );
85 $rdg_text = $rdg->text;
86 # Get the rank from any real reading; they should be identical.
90 if( defined $rdg_text ) {
91 # Initialize the witness array if we haven't got one yet
92 $rdg_wits->{$rdg_text} = [] unless $rdg_wits->{$rdg_text};
93 # Add the relevant witness, subject to a.c. logic
94 add_variant_wit( $rdg_wits->{$rdg_text}, $col_wits->[$j],
95 $tradition->collation->ac_label );
99 # See if this column has any potentially genealogical variants.
100 # If not, skip to the next.
101 $total++ unless scalar keys %$rdg_wits == 1;
102 my( $groups, $readings ) = useful_variant( $rdg_wits );
103 next unless $groups && $readings;
106 # Keep track of our widest row
107 $html_columns = scalar @$groups if scalar @$groups > $html_columns;
109 # We can already look up witnesses for a reading; we also want to look
110 # up readings for a given witness.
111 my $group_readings = {};
112 foreach my $x ( 0 .. $#$groups ) {
113 $group_readings->{wit_stringify( $groups->[$x] )} = $readings->[$x];
116 # For all the groups with more than one member, collect the list of all
117 # contiguous vertices needed to connect them.
118 my $variant_row = analyze_variant_location( $group_readings, $groups,
119 $stemma->graph, $lacunose );
120 $variant_row->{'id'} = $rank;
121 $genealogical++ if $variant_row->{'genealogical'};
122 $conflicts += grep { $_->{'conflict'} } @{$variant_row->{'readings'}};
124 # Now run the same analysis given the calculated distance tree(s).
125 # my @trees = @{$stemma->distance_trees};
127 # foreach my $tree ( 0 .. $#trees ) {
128 # my $dc = analyze_variant_location( $group_readings, $groups, $tree, $lacunose, 'undirected' );
129 # foreach my $rdg ( keys %$dc ) {
130 # my $var = $dc->{$rdg};
131 # # TODO Do something with this
136 # Record that we used this variant in an analysis
137 push( @$variants, $variant_row );
139 my $t1 = Benchmark->new();
140 print STDERR "Analysis of graph for " . $tradition->name . " took "
141 . timestr( timediff( $t1, $t0 ) ) . "seconds\n";
143 # Go through our variant rows, after we have seen all of them once,
144 # and add the number of empty columns needed by each.
145 foreach my $row ( @$variants ) {
146 my $empty = $html_columns - scalar @{$row->{'readings'}};
147 $row->{'empty'} = $empty;
150 # Populate self with our analysis data.
151 $data->{'variants'} = $variants;
152 $data->{'variant_count'} = $total;
153 $data->{'conflict_count'} = $conflicts;
154 $data->{'genealogical_count'} = $genealogical;
155 push( @{$self->{'data'}}, $data );
159 my( $c, $wits ) = @_;
160 my $variant_groups = [];
162 my $all_wits_table = $c->make_alignment_table( 'refs', $wits );
163 # Strip the list of sigla and save it for correlation to the readings.
164 my $col_wits = shift @$all_wits_table;
165 # Any witness in the stemma that has no row should be noted.
166 foreach ( @$col_wits ) {
167 $wits->{$_}++; # Witnesses present in table and stemma now have value 2.
169 my @not_collated = grep { $wits->{$_} == 1 } keys %$wits;
170 foreach my $i ( 0 .. $#$all_wits_table ) {
171 # For each column in the table, group the readings by witness.
173 my $col_rdgs = shift @$all_wits_table;
175 my $lacunose = [ @not_collated ];
176 foreach my $j ( 0 .. $#{$col_rdgs} ) {
177 my $rdg = $col_rdgs->[$j];
178 my $rdg_text = '(omitted)'; # Initialize in case of empty reading
180 if( $rdg->is_lacuna ) {
181 $rdg_text = undef; # Don't count lacunae
182 push( @$lacunose, $col_wits->[$j] );
184 $rdg_text = $rdg->text;
185 # Get the rank from any real reading; they should be identical.
189 if( defined $rdg_text ) {
190 # Initialize the witness array if we haven't got one yet
191 $rdg_wits->{$rdg_text} = [] unless $rdg_wits->{$rdg_text};
192 # Add the relevant witness, subject to a.c. logic
193 add_variant_wit( $rdg_wits->{$rdg_text}, $col_wits->[$j],
198 # See if this column has any potentially genealogical variants.
199 # If not, skip to the next.
200 my( $groups, $readings ) = useful_variant( $rdg_wits );
201 next unless $groups && $readings;
203 push( @$variant_groups, $groups );
205 return $variant_groups;
208 # variant_row -> genealogical
209 # -> readings [ { text, group, conflict, missing } ]
211 sub analyze_variant_location {
212 my( $group_readings, $groups, $graph, $lacunose, $undirected ) = @_;
218 map { $missing->{$_} = 1 } @$lacunose;
219 my $variant_row = { 'readings' => [] };
220 # Mark each ms as in its own group, first.
221 foreach my $g ( @$groups ) {
222 my $gst = wit_stringify( $g );
223 map { $contig->{$_} = $gst } @$g;
225 # Now for each unmarked node in the graph, initialize an array
226 # for possible group memberships. We will use this later to
227 # resolve potential conflicts.
228 map { $contig->{$_} = [] unless $contig->{$_} } $graph->vertices;
229 foreach my $g ( sort { scalar @$b <=> scalar @$a } @$groups ) {
230 my $gst = wit_stringify( $g ); # This is the group name
231 my $reachable = { $g->[0] => 1 };
232 # Copy the graph, and delete all non-members from the new graph.
233 my $part = $graph->copy;
235 $part->delete_vertices(
236 grep { !ref( $contig->{$_} ) && $contig->{$_} ne $gst } $graph->vertices );
238 # Now look to see if our group is connected.
239 if( $undirected ) { # For use with distance trees etc.
240 # Find all vertices reachable from the first (arbitrary) group
241 # member. If we are genealogical this should include them all.
242 map { $reachable->{$_} = 1 } $part->all_reachable( $g->[0] );
243 # TODO This is a terrible way to do distance trees, since all
244 # non-leaf nodes are included in every graph part now. We may
245 # have to go back to SPDP.
248 # Dispense with the trivial case of one reading.
249 # We have to take directionality into account.
250 # How many root nodes do we have?
251 my @roots = grep { ref( $contig->{$_} ) || $contig->{$_} eq $gst }
252 $part->source_vertices;
253 # Assuming that @$g > 1, find the first root node that has at
254 # least one successor belonging to our group. If this reading
255 # is genealogical, there should be only one, but we will check
256 # that implicitly later.
257 my $nodes_in_subtree = 0;
258 foreach my $root ( @roots ) {
259 # Prune the tree to get rid of extraneous hypotheticals.
260 $root = prune_subtree( $part, $root, $contig );
261 # Get all the successor nodes of our root.
262 my $tmp_reach = { $root => 1 };
263 map { $tmp_reach->{$_} = 1 } $part->all_successors( $root );
264 # Skip this root if none of our successors are in our group
265 # (e.g. isolated 'hypothetical' witnesses with no group)
266 next unless grep { $contig->{$_} } keys %$tmp_reach;
267 if( keys %$tmp_reach > $nodes_in_subtree ) {
268 $nodes_in_subtree = keys %$tmp_reach;
269 $reachable = $tmp_reach;
273 } # else it is a single-node group, nothing to calculate.
276 # None of the 'reachable' nodes should be marked as being in another
277 # group. Paint the 'hypotheticals' with our group while we are at it,
278 # unless there is a conflict present.
279 foreach ( keys %$reachable ) {
280 if( ref $contig->{$_} ) {
281 push( @{$contig->{$_}}, $gst );
282 } elsif( $contig->{$_} ne $gst ) {
283 $conflict->{$group_readings->{$gst}} = $group_readings->{$contig->{$_}};
284 } # else it is an 'extant' node marked with our group already.
286 # None of the unreachable nodes should be in our group either.
287 foreach ( $part->vertices ) {
288 next if $reachable->{$_};
289 if( $contig->{$_} eq $gst ) {
290 $conflict->{$group_readings->{$gst}} = $group_readings->{$gst};
295 # Now, if we have a conflict, we can write the reading in full. If not,
296 # we have to save the subgraph so that we can resolve possible conflicts
297 # on hypothetical nodes.
298 $is_conflicted = 1 if exists $conflict->{$group_readings->{$gst}};
301 my $reading = { 'text' => $group_readings->{$gst},
302 'missing' => wit_stringify( $lacunose ),
303 'group' => $gst }; # This will change if we find no conflict
304 if( $is_conflicted ) {
305 $reading->{'conflict'} = $conflict->{$group_readings->{$gst}}
307 # Save the relevant subgraph.
308 $subgraph->{$gst} = { 'graph' => $part,
309 'root' => $group_root,
310 'reachable' => $reachable };
312 push( @{$variant_row->{'readings'}}, $reading );
315 # Now that we have gone through all the rows, check the hypothetical
316 # readings for conflict if we haven't found one yet.
317 if( keys %$subgraph && !keys %$conflict ) {
319 foreach ( keys %$contig ) {
320 next unless ref $contig->{$_};
321 if( scalar @{$contig->{$_}} > 1 ) {
322 push( @resolve, $_ );
324 $contig->{$_} = scalar @{$contig->{$_}} ? $contig->{$_}->[0] : '';
327 # Do we still have a possible conflict?
328 my $still_contig = {};
329 foreach my $h ( @resolve ) {
330 # For each of the hypothetical readings with more than one possibility,
331 # try deleting it from each of its member subgraphs in turn, and see
332 # if that breaks the contiguous grouping.
333 # TODO This can still break in a corner case where group A can use
334 # either vertex 1 or 2, and group B can use either vertex 2 or 1.
335 # Revisit this if necessary; it could get brute-force nasty.
336 foreach my $gst ( @{$contig->{$h}} ) {
337 my $gpart = $subgraph->{$gst}->{'graph'}->copy;
338 my $reachable = $subgraph->{$gst}->{'reachable'};
339 $gpart->delete_vertex( $h );
340 # Is everything else still reachable from the root?
341 # TODO If $h was the root, see if we still have a single root.
342 my %still_reachable = ( $subgraph->{$gst}->{'root'} => 1 );
343 map { $still_reachable{$_} = 1 }
344 $gpart->all_successors( $subgraph->{$gst}->{'root'} );
345 foreach my $v ( keys %$reachable ) {
347 if( !$still_reachable{$v}
348 && ( $contig->{$v} eq $gst
349 || ( exists $still_contig->{$v}
350 && $still_contig->{$v} eq $gst ) ) ) {
352 if( exists $still_contig->{$h} ) {
354 $conflict->{$group_readings->{$gst}} =
355 $group_readings->{$still_contig->{$h}};
357 $still_contig->{$h} = $gst;
360 } # else we don't need $h in this group.
365 # Now, assuming no conflict, we have some hypothetical vertices in
366 # $still_contig that are the "real" group memberships. Replace these
368 unless ( keys %$conflict ) {
369 foreach my $v ( keys %$contig ) {
370 next unless ref $contig->{$v};
371 $contig->{$v} = $still_contig->{$v};
376 # Now write the group and conflict information into the respective rows.
377 foreach my $rdg ( @{$variant_row->{'readings'}} ) {
378 $rdg->{'conflict'} = $conflict->{$rdg->{'text'}};
379 next if $rdg->{'conflict'};
380 my @members = grep { $contig->{$_} eq $rdg->{'group'} && !$missing->{$_} }
382 $rdg->{'group'} = wit_stringify( \@members );
385 $variant_row->{'genealogical'} = !( keys %$conflict );
390 my( $tree, $root, $contighash ) = @_;
391 # First, delete hypothetical leaves / orphans until there are none left.
392 my @orphan_hypotheticals = grep { ref( $contighash->{$_} ) }
393 $tree->successorless_vertices;
394 while( @orphan_hypotheticals ) {
395 $tree->delete_vertices( @orphan_hypotheticals );
396 @orphan_hypotheticals = grep { ref( $contighash->{$_} ) }
397 $tree->successorless_vertices;
399 # Then delete a hypothetical root with only one successor, moving the
401 while( $tree->successors( $root ) == 1 && ref $contighash->{$root} ) {
402 my @nextroot = $tree->successors( $root );
403 $tree->delete_vertex( $root );
404 $root = $nextroot[0];
406 # The tree has been modified in place, but we need to know the new root.
409 # Add the variant, subject to a.c. representation logic.
410 # This assumes that we will see the 'main' version before the a.c. version.
411 sub add_variant_wit {
412 my( $arr, $wit, $acstr ) = @_;
414 if( $wit =~ /^(.*)\Q$acstr\E$/ ) {
416 $skip = grep { $_ =~ /^\Q$real\E$/ } @$arr;
418 push( @$arr, $wit ) unless $skip;
421 # Return an answer if the variant is useful, i.e. if there are at least 2 variants
422 # with at least 2 witnesses each.
424 my( $readings ) = @_;
425 my $total = keys %$readings;
426 foreach my $var ( keys %$readings ) {
427 $total-- if @{$readings->{$var}} == 1;
429 return( undef, undef ) if $total <= 1;
430 my( $groups, $text );
431 foreach my $var ( keys %$readings ) {
432 push( @$groups, $readings->{$var} );
433 push( @$text, $var );
435 return( $groups, $text );
438 # Take an array of witness groupings and produce a string like
439 # ['A','B'] / ['C','D','E'] / ['F']
444 # If we were passed an array of witnesses instead of an array of
445 # groupings, then "group" the witnesses first.
446 unless( ref( $groups->[0] ) ) {
447 my $mkgrp = [ $groups ];
450 foreach my $g ( @$groups ) {
451 push( @gst, '[' . join( ',', map { "'$_'" } @$g ) . ']' );
453 return join( ' / ', @gst );