Commit | Line | Data |
d71100ed |
1 | package Text::Tradition::Analysis; |
2 | |
3 | use strict; |
4 | use warnings; |
e4386ba9 |
5 | use Benchmark; |
d1348d38 |
6 | use Exporter 'import'; |
d71100ed |
7 | use Text::Tradition; |
8 | use Text::Tradition::Stemma; |
9 | |
d1348d38 |
10 | use vars qw/ @EXPORT_OK /; |
11 | @EXPORT_OK = qw/ run_analysis group_variants wit_stringify /; |
12 | |
d71100ed |
13 | sub run_analysis { |
3837c155 |
14 | my( $tradition, $stemma ) = @_; |
d71100ed |
15 | # What we will return |
d71100ed |
16 | my $variants = []; |
e367f5c0 |
17 | my $data = {}; |
3837c155 |
18 | |
d71100ed |
19 | # We have the collation, so get the alignment table with witnesses in rows. |
20 | # Also return the reading objects in the table, rather than just the words. |
08e0fb85 |
21 | my $wits = {}; |
22 | map { $wits->{$_} = 1 } $stemma->witnesses; |
d71100ed |
23 | # For each column in the alignment table, we want to see if the existing |
f6c8ea08 |
24 | # groupings of witnesses match our stemma hypothesis. We also need to keep |
25 | # track of the maximum number of variants at any one location. |
26 | my $max_variants = 0; |
732152b1 |
27 | my ( $total, $genealogical, $conflicts ) = ( 0, 0, 0 ); |
d71100ed |
28 | |
e4386ba9 |
29 | my $t0 = Benchmark->new(); |
f6c8ea08 |
30 | my $variant_groups = group_variants( $tradition->collation, $wits ); |
31 | foreach my $rank ( 0 .. $#{$variant_groups} ) { |
32 | my $groups = $variant_groups->[$rank]->{'groups'}; |
33 | my $readings = $variant_groups->[$rank]->{'readings'}; |
34 | my $lacunose = $variant_groups->[$rank]->{'lacunose'}; |
d71100ed |
35 | |
f6c8ea08 |
36 | $max_variants = scalar @$groups if scalar @$groups > $max_variants; |
d71100ed |
37 | |
38 | # We can already look up witnesses for a reading; we also want to look |
39 | # up readings for a given witness. |
40 | my $group_readings = {}; |
41 | foreach my $x ( 0 .. $#$groups ) { |
42 | $group_readings->{wit_stringify( $groups->[$x] )} = $readings->[$x]; |
43 | } |
44 | |
45 | # For all the groups with more than one member, collect the list of all |
46 | # contiguous vertices needed to connect them. |
f6c8ea08 |
47 | my $variant_loc = analyze_variant_location( $group_readings, $groups, |
08e0fb85 |
48 | $stemma->graph, $lacunose ); |
f6c8ea08 |
49 | $variant_loc->{'id'} = $rank; |
50 | $genealogical++ if $variant_loc->{'genealogical'}; |
51 | $conflicts += grep { $_->{'conflict'} } @{$variant_loc->{'readings'}}; |
732152b1 |
52 | |
d71100ed |
53 | # Now run the same analysis given the calculated distance tree(s). |
732152b1 |
54 | # my @trees = @{$stemma->distance_trees}; |
55 | # if( @trees ) { |
56 | # foreach my $tree ( 0 .. $#trees ) { |
c4a4fb1b |
57 | # my $dc = analyze_variant_location( $group_readings, $groups, $tree, $lacunose, 'undirected' ); |
732152b1 |
58 | # foreach my $rdg ( keys %$dc ) { |
59 | # my $var = $dc->{$rdg}; |
60 | # # TODO Do something with this |
61 | # } |
62 | # } |
63 | # } |
64 | |
d71100ed |
65 | # Record that we used this variant in an analysis |
f6c8ea08 |
66 | push( @$variants, $variant_loc ); |
d71100ed |
67 | } |
e4386ba9 |
68 | my $t1 = Benchmark->new(); |
69 | print STDERR "Analysis of graph for " . $tradition->name . " took " |
70 | . timestr( timediff( $t1, $t0 ) ) . "seconds\n"; |
d71100ed |
71 | |
f6c8ea08 |
72 | # Go through our variant locations, after we have seen all of them once, |
732152b1 |
73 | # and add the number of empty columns needed by each. |
d71100ed |
74 | foreach my $row ( @$variants ) { |
f6c8ea08 |
75 | my $empty = $max_variants - scalar @{$row->{'readings'}}; |
d71100ed |
76 | $row->{'empty'} = $empty; |
77 | } |
78 | |
e367f5c0 |
79 | $data->{'variants'} = $variants; |
80 | $data->{'variant_count'} = $total; |
81 | $data->{'conflict_count'} = $conflicts; |
82 | $data->{'genealogical_count'} = $genealogical; |
3837c155 |
83 | return $data; |
d71100ed |
84 | } |
85 | |
d1348d38 |
86 | sub group_variants { |
87 | my( $c, $wits ) = @_; |
88 | my $variant_groups = []; |
89 | |
f6c8ea08 |
90 | # We have the collation, so get the alignment table with witnesses in rows. |
91 | # Also return the reading objects in the table, rather than just the words. |
d1348d38 |
92 | my $all_wits_table = $c->make_alignment_table( 'refs', $wits ); |
93 | # Strip the list of sigla and save it for correlation to the readings. |
94 | my $col_wits = shift @$all_wits_table; |
95 | # Any witness in the stemma that has no row should be noted. |
96 | foreach ( @$col_wits ) { |
97 | $wits->{$_}++; # Witnesses present in table and stemma now have value 2. |
98 | } |
f6c8ea08 |
99 | my @not_collated = grep { $wits->{$_} == 1 } keys %$wits; |
d1348d38 |
100 | foreach my $i ( 0 .. $#$all_wits_table ) { |
101 | # For each column in the table, group the readings by witness. |
102 | my $rdg_wits = {}; |
103 | my $col_rdgs = shift @$all_wits_table; |
d1348d38 |
104 | my $lacunose = [ @not_collated ]; |
105 | foreach my $j ( 0 .. $#{$col_rdgs} ) { |
106 | my $rdg = $col_rdgs->[$j]; |
107 | my $rdg_text = '(omitted)'; # Initialize in case of empty reading |
108 | if( $rdg ) { |
109 | if( $rdg->is_lacuna ) { |
110 | $rdg_text = undef; # Don't count lacunae |
111 | push( @$lacunose, $col_wits->[$j] ); |
112 | } else { |
113 | $rdg_text = $rdg->text; |
d1348d38 |
114 | } |
115 | } |
116 | if( defined $rdg_text ) { |
117 | # Initialize the witness array if we haven't got one yet |
118 | $rdg_wits->{$rdg_text} = [] unless $rdg_wits->{$rdg_text}; |
119 | # Add the relevant witness, subject to a.c. logic |
120 | add_variant_wit( $rdg_wits->{$rdg_text}, $col_wits->[$j], |
121 | $c->ac_label ); |
122 | } |
123 | } |
124 | |
125 | # See if this column has any potentially genealogical variants. |
126 | # If not, skip to the next. |
127 | my( $groups, $readings ) = useful_variant( $rdg_wits ); |
128 | next unless $groups && $readings; |
129 | |
f6c8ea08 |
130 | push( @$variant_groups, |
131 | { 'groups' => $groups, 'readings' => $readings, 'lacunose' => $lacunose } ); |
d1348d38 |
132 | } |
133 | return $variant_groups; |
134 | } |
135 | |
f6c8ea08 |
136 | |
137 | |
732152b1 |
138 | # variant_row -> genealogical |
139 | # -> readings [ { text, group, conflict, missing } ] |
140 | |
d71100ed |
141 | sub analyze_variant_location { |
c4a4fb1b |
142 | my( $group_readings, $groups, $graph, $lacunose, $undirected ) = @_; |
231d71fc |
143 | my $contig = {}; |
144 | my $subgraph = {}; |
c4a4fb1b |
145 | my $is_conflicted; |
d71100ed |
146 | my $conflict = {}; |
231d71fc |
147 | my $missing = {}; |
148 | map { $missing->{$_} = 1 } @$lacunose; |
732152b1 |
149 | my $variant_row = { 'readings' => [] }; |
94a077d6 |
150 | # Mark each ms as in its own group, first. |
151 | foreach my $g ( @$groups ) { |
152 | my $gst = wit_stringify( $g ); |
231d71fc |
153 | map { $contig->{$_} = $gst } @$g; |
94a077d6 |
154 | } |
c4a4fb1b |
155 | # Now for each unmarked node in the graph, initialize an array |
156 | # for possible group memberships. We will use this later to |
157 | # resolve potential conflicts. |
231d71fc |
158 | map { $contig->{$_} = [] unless $contig->{$_} } $graph->vertices; |
d71100ed |
159 | foreach my $g ( sort { scalar @$b <=> scalar @$a } @$groups ) { |
c4a4fb1b |
160 | my $gst = wit_stringify( $g ); # This is the group name |
161 | my $reachable = { $g->[0] => 1 }; |
08e0fb85 |
162 | # Copy the graph, and delete all non-members from the new graph. |
c4a4fb1b |
163 | my $part = $graph->copy; |
164 | my $group_root; |
165 | $part->delete_vertices( |
231d71fc |
166 | grep { !ref( $contig->{$_} ) && $contig->{$_} ne $gst } $graph->vertices ); |
c4a4fb1b |
167 | |
168 | # Now look to see if our group is connected. |
169 | if( $undirected ) { # For use with distance trees etc. |
170 | # Find all vertices reachable from the first (arbitrary) group |
171 | # member. If we are genealogical this should include them all. |
172 | map { $reachable->{$_} = 1 } $part->all_reachable( $g->[0] ); |
173 | # TODO This is a terrible way to do distance trees, since all |
174 | # non-leaf nodes are included in every graph part now. We may |
175 | # have to go back to SPDP. |
176 | } else { |
177 | if( @$g > 1 ) { |
178 | # Dispense with the trivial case of one reading. |
179 | # We have to take directionality into account. |
180 | # How many root nodes do we have? |
231d71fc |
181 | my @roots = grep { ref( $contig->{$_} ) || $contig->{$_} eq $gst } |
c4a4fb1b |
182 | $part->source_vertices; |
183 | # Assuming that @$g > 1, find the first root node that has at |
184 | # least one successor belonging to our group. If this reading |
185 | # is genealogical, there should be only one, but we will check |
186 | # that implicitly later. |
187 | my $nodes_in_subtree = 0; |
188 | foreach my $root ( @roots ) { |
231d71fc |
189 | # Prune the tree to get rid of extraneous hypotheticals. |
190 | $root = prune_subtree( $part, $root, $contig ); |
c4a4fb1b |
191 | # Get all the successor nodes of our root. |
192 | my $tmp_reach = { $root => 1 }; |
193 | map { $tmp_reach->{$_} = 1 } $part->all_successors( $root ); |
194 | # Skip this root if none of our successors are in our group |
195 | # (e.g. isolated 'hypothetical' witnesses with no group) |
231d71fc |
196 | next unless grep { $contig->{$_} } keys %$tmp_reach; |
c4a4fb1b |
197 | if( keys %$tmp_reach > $nodes_in_subtree ) { |
198 | $nodes_in_subtree = keys %$tmp_reach; |
199 | $reachable = $tmp_reach; |
200 | $group_root = $root; |
201 | } |
202 | } |
203 | } # else it is a single-node group, nothing to calculate. |
204 | } |
205 | |
206 | # None of the 'reachable' nodes should be marked as being in another |
207 | # group. Paint the 'hypotheticals' with our group while we are at it, |
208 | # unless there is a conflict present. |
209 | foreach ( keys %$reachable ) { |
231d71fc |
210 | if( ref $contig->{$_} ) { |
211 | push( @{$contig->{$_}}, $gst ); |
212 | } elsif( $contig->{$_} ne $gst ) { |
213 | $conflict->{$group_readings->{$gst}} = $group_readings->{$contig->{$_}}; |
c4a4fb1b |
214 | } # else it is an 'extant' node marked with our group already. |
d71100ed |
215 | } |
08e0fb85 |
216 | # None of the unreachable nodes should be in our group either. |
217 | foreach ( $part->vertices ) { |
c4a4fb1b |
218 | next if $reachable->{$_}; |
231d71fc |
219 | if( $contig->{$_} eq $gst ) { |
c4a4fb1b |
220 | $conflict->{$group_readings->{$gst}} = $group_readings->{$gst}; |
221 | last; |
222 | } |
08e0fb85 |
223 | } |
224 | |
c4a4fb1b |
225 | # Now, if we have a conflict, we can write the reading in full. If not, |
226 | # we have to save the subgraph so that we can resolve possible conflicts |
227 | # on hypothetical nodes. |
228 | $is_conflicted = 1 if exists $conflict->{$group_readings->{$gst}}; |
229 | |
732152b1 |
230 | # Write the reading. |
231 | my $reading = { 'text' => $group_readings->{$gst}, |
232 | 'missing' => wit_stringify( $lacunose ), |
c4a4fb1b |
233 | 'group' => $gst }; # This will change if we find no conflict |
234 | if( $is_conflicted ) { |
235 | $reading->{'conflict'} = $conflict->{$group_readings->{$gst}} |
732152b1 |
236 | } else { |
c4a4fb1b |
237 | # Save the relevant subgraph. |
231d71fc |
238 | $subgraph->{$gst} = { 'graph' => $part, |
c4a4fb1b |
239 | 'root' => $group_root, |
240 | 'reachable' => $reachable }; |
732152b1 |
241 | } |
242 | push( @{$variant_row->{'readings'}}, $reading ); |
d71100ed |
243 | } |
c4a4fb1b |
244 | |
245 | # Now that we have gone through all the rows, check the hypothetical |
246 | # readings for conflict if we haven't found one yet. |
231d71fc |
247 | if( keys %$subgraph && !keys %$conflict ) { |
c4a4fb1b |
248 | my @resolve; |
231d71fc |
249 | foreach ( keys %$contig ) { |
250 | next unless ref $contig->{$_}; |
251 | if( scalar @{$contig->{$_}} > 1 ) { |
c4a4fb1b |
252 | push( @resolve, $_ ); |
253 | } else { |
231d71fc |
254 | $contig->{$_} = scalar @{$contig->{$_}} ? $contig->{$_}->[0] : ''; |
c4a4fb1b |
255 | } |
256 | } |
257 | # Do we still have a possible conflict? |
231d71fc |
258 | my $still_contig = {}; |
c4a4fb1b |
259 | foreach my $h ( @resolve ) { |
260 | # For each of the hypothetical readings with more than one possibility, |
261 | # try deleting it from each of its member subgraphs in turn, and see |
262 | # if that breaks the contiguous grouping. |
263 | # TODO This can still break in a corner case where group A can use |
264 | # either vertex 1 or 2, and group B can use either vertex 2 or 1. |
265 | # Revisit this if necessary; it could get brute-force nasty. |
231d71fc |
266 | foreach my $gst ( @{$contig->{$h}} ) { |
267 | my $gpart = $subgraph->{$gst}->{'graph'}->copy; |
268 | my $reachable = $subgraph->{$gst}->{'reachable'}; |
c4a4fb1b |
269 | $gpart->delete_vertex( $h ); |
270 | # Is everything else still reachable from the root? |
271 | # TODO If $h was the root, see if we still have a single root. |
231d71fc |
272 | my %still_reachable = ( $subgraph->{$gst}->{'root'} => 1 ); |
c4a4fb1b |
273 | map { $still_reachable{$_} = 1 } |
231d71fc |
274 | $gpart->all_successors( $subgraph->{$gst}->{'root'} ); |
c4a4fb1b |
275 | foreach my $v ( keys %$reachable ) { |
276 | next if $v eq $h; |
277 | if( !$still_reachable{$v} |
231d71fc |
278 | && ( $contig->{$v} eq $gst |
279 | || ( exists $still_contig->{$v} |
280 | && $still_contig->{$v} eq $gst ) ) ) { |
c4a4fb1b |
281 | # We need $h. |
231d71fc |
282 | if( exists $still_contig->{$h} ) { |
c4a4fb1b |
283 | # Conflict! |
284 | $conflict->{$group_readings->{$gst}} = |
231d71fc |
285 | $group_readings->{$still_contig->{$h}}; |
c4a4fb1b |
286 | } else { |
231d71fc |
287 | $still_contig->{$h} = $gst; |
c4a4fb1b |
288 | } |
289 | last; |
290 | } # else we don't need $h in this group. |
291 | } |
292 | } |
293 | } |
294 | |
295 | # Now, assuming no conflict, we have some hypothetical vertices in |
296 | # $still_contig that are the "real" group memberships. Replace these |
297 | # in $contig. |
298 | unless ( keys %$conflict ) { |
231d71fc |
299 | foreach my $v ( keys %$contig ) { |
300 | next unless ref $contig->{$v}; |
301 | $contig->{$v} = $still_contig->{$v}; |
c4a4fb1b |
302 | } |
303 | } |
304 | } |
305 | |
306 | # Now write the group and conflict information into the respective rows. |
307 | foreach my $rdg ( @{$variant_row->{'readings'}} ) { |
308 | $rdg->{'conflict'} = $conflict->{$rdg->{'text'}}; |
309 | next if $rdg->{'conflict'}; |
231d71fc |
310 | my @members = grep { $contig->{$_} eq $rdg->{'group'} && !$missing->{$_} } |
311 | keys %$contig; |
c4a4fb1b |
312 | $rdg->{'group'} = wit_stringify( \@members ); |
313 | } |
314 | |
08e0fb85 |
315 | $variant_row->{'genealogical'} = !( keys %$conflict ); |
732152b1 |
316 | return $variant_row; |
d71100ed |
317 | } |
318 | |
231d71fc |
319 | sub prune_subtree { |
320 | my( $tree, $root, $contighash ) = @_; |
321 | # First, delete hypothetical leaves / orphans until there are none left. |
322 | my @orphan_hypotheticals = grep { ref( $contighash->{$_} ) } |
323 | $tree->successorless_vertices; |
324 | while( @orphan_hypotheticals ) { |
325 | $tree->delete_vertices( @orphan_hypotheticals ); |
326 | @orphan_hypotheticals = grep { ref( $contighash->{$_} ) } |
327 | $tree->successorless_vertices; |
328 | } |
329 | # Then delete a hypothetical root with only one successor, moving the |
330 | # root to the child. |
331 | while( $tree->successors( $root ) == 1 && ref $contighash->{$root} ) { |
332 | my @nextroot = $tree->successors( $root ); |
333 | $tree->delete_vertex( $root ); |
334 | $root = $nextroot[0]; |
335 | } |
336 | # The tree has been modified in place, but we need to know the new root. |
337 | return $root; |
338 | } |
d71100ed |
339 | # Add the variant, subject to a.c. representation logic. |
340 | # This assumes that we will see the 'main' version before the a.c. version. |
341 | sub add_variant_wit { |
342 | my( $arr, $wit, $acstr ) = @_; |
343 | my $skip; |
344 | if( $wit =~ /^(.*)\Q$acstr\E$/ ) { |
345 | my $real = $1; |
346 | $skip = grep { $_ =~ /^\Q$real\E$/ } @$arr; |
347 | } |
348 | push( @$arr, $wit ) unless $skip; |
349 | } |
350 | |
351 | # Return an answer if the variant is useful, i.e. if there are at least 2 variants |
352 | # with at least 2 witnesses each. |
353 | sub useful_variant { |
354 | my( $readings ) = @_; |
355 | my $total = keys %$readings; |
356 | foreach my $var ( keys %$readings ) { |
357 | $total-- if @{$readings->{$var}} == 1; |
358 | } |
359 | return( undef, undef ) if $total <= 1; |
360 | my( $groups, $text ); |
361 | foreach my $var ( keys %$readings ) { |
362 | push( @$groups, $readings->{$var} ); |
363 | push( @$text, $var ); |
364 | } |
365 | return( $groups, $text ); |
366 | } |
367 | |
368 | # Take an array of witness groupings and produce a string like |
369 | # ['A','B'] / ['C','D','E'] / ['F'] |
370 | |
371 | sub wit_stringify { |
372 | my $groups = shift; |
373 | my @gst; |
374 | # If we were passed an array of witnesses instead of an array of |
375 | # groupings, then "group" the witnesses first. |
376 | unless( ref( $groups->[0] ) ) { |
377 | my $mkgrp = [ $groups ]; |
378 | $groups = $mkgrp; |
379 | } |
380 | foreach my $g ( @$groups ) { |
381 | push( @gst, '[' . join( ',', map { "'$_'" } @$g ) . ']' ); |
382 | } |
383 | return join( ' / ', @gst ); |
384 | } |
385 | |
386 | 1; |