Commit | Line | Data |
d71100ed |
1 | package Text::Tradition::Analysis; |
2 | |
3 | use strict; |
4 | use warnings; |
5 | use Text::Tradition; |
6 | use Text::Tradition::Stemma; |
7 | |
8 | sub new { |
9 | my( $class, $args ) = @_; |
10 | my $self = {}; |
d71100ed |
11 | bless( $self, $class ); |
e367f5c0 |
12 | $self->{'data'} = []; |
13 | foreach my $t ( @{$args->{'traditions'}} ) { |
14 | $self->run_analysis( $t->{'file'}, $t->{'stemmadot'} ); |
15 | } |
d71100ed |
16 | return $self; |
17 | } |
18 | |
19 | sub run_analysis { |
732152b1 |
20 | my( $self, $file, $stemmadot ) = @_; |
d71100ed |
21 | # What we will return |
22 | my $svg; |
23 | my $variants = []; |
e367f5c0 |
24 | my $data = {}; |
d71100ed |
25 | |
3d79e248 |
26 | # Read in the file and stemma |
d71100ed |
27 | my $tradition = Text::Tradition->new( |
3d79e248 |
28 | 'input' => 'Self', |
29 | 'file' => $file, |
d71100ed |
30 | 'linear' => 1, |
31 | ); |
e367f5c0 |
32 | $data->{'title'} = $tradition->name; |
732152b1 |
33 | |
d71100ed |
34 | my $stemma = Text::Tradition::Stemma->new( |
35 | 'collation' => $tradition->collation, |
36 | 'dot' => $stemmadot, |
37 | ); |
38 | # We will return the stemma picture |
e367f5c0 |
39 | $svg = $stemma->as_svg( { size => "8,7.5" } );; |
40 | $data->{'svg'} = $svg; |
d71100ed |
41 | |
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. |
08e0fb85 |
44 | my $wits = {}; |
45 | map { $wits->{$_} = 1 } $stemma->witnesses; |
46 | my $all_wits_table = $tradition->collation->make_alignment_table( 'refs', $wits ); |
d71100ed |
47 | |
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. |
51 | my $html_columns = 0; |
732152b1 |
52 | my ( $total, $genealogical, $conflicts ) = ( 0, 0, 0 ); |
d71100ed |
53 | |
54 | # Strip the list of sigla and save it for correlation to the readings. |
55 | my $col_wits = shift @$all_wits_table; |
56 | |
57 | # We will return a data structure, an array for each row that looks like: |
58 | # { id = X, genealogical = Y, readings = [ text = X, group = Y], empty = N } |
59 | foreach my $i ( 0 .. $#$all_wits_table ) { |
60 | # For each column in the table, group the readings by witness. |
61 | my $rdg_wits = {}; |
62 | my $col_rdgs = shift @$all_wits_table; |
63 | my $rank; |
c4e11e3f |
64 | my $lacunose = []; |
d71100ed |
65 | foreach my $j ( 0 .. $#{$col_rdgs} ) { |
66 | my $rdg = $col_rdgs->[$j]; |
d71100ed |
67 | my $rdg_text = '(omitted)'; # Initialize in case of empty reading |
68 | if( $rdg ) { |
c4e11e3f |
69 | if( $rdg->is_lacuna ) { |
70 | $rdg_text = undef; # Don't count lacunae |
71 | push( @$lacunose, $col_wits->[$j] ); |
72 | } else { |
73 | $rdg_text = $rdg->text; |
74 | # Get the rank from any real reading; they should be identical. |
75 | $rank = $rdg->rank; |
76 | } |
d71100ed |
77 | } |
78 | if( defined $rdg_text ) { |
79 | # Initialize the witness array if we haven't got one yet |
80 | $rdg_wits->{$rdg_text} = [] unless $rdg_wits->{$rdg_text}; |
81 | # Add the relevant witness, subject to a.c. logic |
82 | add_variant_wit( $rdg_wits->{$rdg_text}, $col_wits->[$j], |
83 | $tradition->collation->ac_label ); |
84 | } |
85 | } |
86 | |
87 | # See if this column has any potentially genealogical variants. |
88 | # If not, skip to the next. |
89 | $total++ unless scalar keys %$rdg_wits == 1; |
90 | my( $groups, $readings ) = useful_variant( $rdg_wits ); |
91 | next unless $groups && $readings; |
92 | |
d71100ed |
93 | # Keep track of our widest row |
94 | $html_columns = scalar @$groups if scalar @$groups > $html_columns; |
95 | |
96 | # We can already look up witnesses for a reading; we also want to look |
97 | # up readings for a given witness. |
98 | my $group_readings = {}; |
99 | foreach my $x ( 0 .. $#$groups ) { |
100 | $group_readings->{wit_stringify( $groups->[$x] )} = $readings->[$x]; |
101 | } |
102 | |
103 | # For all the groups with more than one member, collect the list of all |
104 | # contiguous vertices needed to connect them. |
c4a4fb1b |
105 | $DB::single = 1 if $rank == 733; |
732152b1 |
106 | my $variant_row = analyze_variant_location( $group_readings, $groups, |
08e0fb85 |
107 | $stemma->graph, $lacunose ); |
732152b1 |
108 | $variant_row->{'id'} = $rank; |
109 | $genealogical++ if $variant_row->{'genealogical'}; |
110 | $conflicts += grep { $_->{'conflict'} } @{$variant_row->{'readings'}}; |
111 | |
d71100ed |
112 | # Now run the same analysis given the calculated distance tree(s). |
732152b1 |
113 | # my @trees = @{$stemma->distance_trees}; |
114 | # if( @trees ) { |
115 | # foreach my $tree ( 0 .. $#trees ) { |
c4a4fb1b |
116 | # my $dc = analyze_variant_location( $group_readings, $groups, $tree, $lacunose, 'undirected' ); |
732152b1 |
117 | # foreach my $rdg ( keys %$dc ) { |
118 | # my $var = $dc->{$rdg}; |
119 | # # TODO Do something with this |
120 | # } |
121 | # } |
122 | # } |
123 | |
d71100ed |
124 | # Record that we used this variant in an analysis |
125 | push( @$variants, $variant_row ); |
126 | } |
127 | |
732152b1 |
128 | # Go through our variant rows, after we have seen all of them once, |
129 | # and add the number of empty columns needed by each. |
d71100ed |
130 | foreach my $row ( @$variants ) { |
131 | my $empty = $html_columns - scalar @{$row->{'readings'}}; |
132 | $row->{'empty'} = $empty; |
133 | } |
134 | |
732152b1 |
135 | # Populate self with our analysis data. |
e367f5c0 |
136 | $data->{'variants'} = $variants; |
137 | $data->{'variant_count'} = $total; |
138 | $data->{'conflict_count'} = $conflicts; |
139 | $data->{'genealogical_count'} = $genealogical; |
140 | push( @{$self->{'data'}}, $data ); |
d71100ed |
141 | } |
142 | |
732152b1 |
143 | # variant_row -> genealogical |
144 | # -> readings [ { text, group, conflict, missing } ] |
145 | |
d71100ed |
146 | sub analyze_variant_location { |
c4a4fb1b |
147 | my( $group_readings, $groups, $graph, $lacunose, $undirected ) = @_; |
d71100ed |
148 | my %contig; |
c4a4fb1b |
149 | my %subgraph; |
150 | my $is_conflicted; |
d71100ed |
151 | my $conflict = {}; |
732152b1 |
152 | my %missing; |
153 | map { $missing{$_} = 1 } @$lacunose; |
154 | my $variant_row = { 'readings' => [] }; |
94a077d6 |
155 | # Mark each ms as in its own group, first. |
156 | foreach my $g ( @$groups ) { |
157 | my $gst = wit_stringify( $g ); |
158 | map { $contig{$_} = $gst } @$g; |
159 | } |
c4a4fb1b |
160 | # Now for each unmarked node in the graph, initialize an array |
161 | # for possible group memberships. We will use this later to |
162 | # resolve potential conflicts. |
163 | map { $contig{$_} = [] unless $contig{$_} } $graph->vertices; |
d71100ed |
164 | foreach my $g ( sort { scalar @$b <=> scalar @$a } @$groups ) { |
c4a4fb1b |
165 | my $gst = wit_stringify( $g ); # This is the group name |
166 | my $reachable = { $g->[0] => 1 }; |
08e0fb85 |
167 | # Copy the graph, and delete all non-members from the new graph. |
c4a4fb1b |
168 | my $part = $graph->copy; |
169 | my $group_root; |
170 | $part->delete_vertices( |
171 | grep { !ref( $contig{$_} ) && $contig{$_} ne $gst } $graph->vertices ); |
172 | |
173 | # Now look to see if our group is connected. |
174 | if( $undirected ) { # For use with distance trees etc. |
175 | # Find all vertices reachable from the first (arbitrary) group |
176 | # member. If we are genealogical this should include them all. |
177 | map { $reachable->{$_} = 1 } $part->all_reachable( $g->[0] ); |
178 | # TODO This is a terrible way to do distance trees, since all |
179 | # non-leaf nodes are included in every graph part now. We may |
180 | # have to go back to SPDP. |
181 | } else { |
182 | if( @$g > 1 ) { |
183 | # Dispense with the trivial case of one reading. |
184 | # We have to take directionality into account. |
185 | # How many root nodes do we have? |
186 | my @roots = grep { ref( $contig{$_} ) || $contig{$_} eq $gst } |
187 | $part->source_vertices; |
188 | # Assuming that @$g > 1, find the first root node that has at |
189 | # least one successor belonging to our group. If this reading |
190 | # is genealogical, there should be only one, but we will check |
191 | # that implicitly later. |
192 | my $nodes_in_subtree = 0; |
193 | foreach my $root ( @roots ) { |
194 | # Prune the tree down to the first root that either is extant |
195 | # or has more than one successor. |
196 | while( $part->successors( $root ) == 1 && ref $contig{$root} ) { |
197 | my @nextroot = $part->successors( $root ); |
198 | $part->delete_vertex( $root ); |
199 | $root = $nextroot[0]; |
200 | } |
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; |
210 | $group_root = $root; |
211 | } |
212 | } |
213 | } # else it is a single-node group, nothing to calculate. |
214 | } |
215 | |
216 | # None of the 'reachable' nodes should be marked as being in another |
217 | # group. Paint the 'hypotheticals' with our group while we are at it, |
218 | # unless there is a conflict present. |
219 | foreach ( keys %$reachable ) { |
220 | if( ref $contig{$_} ) { |
221 | push( @{$contig{$_}}, $gst ); |
222 | } elsif( $contig{$_} ne $gst ) { |
08e0fb85 |
223 | $conflict->{$group_readings->{$gst}} = $group_readings->{$contig{$_}}; |
c4a4fb1b |
224 | } # else it is an 'extant' node marked with our group already. |
d71100ed |
225 | } |
08e0fb85 |
226 | # None of the unreachable nodes should be in our group either. |
227 | foreach ( $part->vertices ) { |
c4a4fb1b |
228 | next if $reachable->{$_}; |
229 | if( $contig{$_} eq $gst ) { |
230 | $conflict->{$group_readings->{$gst}} = $group_readings->{$gst}; |
231 | last; |
232 | } |
08e0fb85 |
233 | } |
234 | |
c4a4fb1b |
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}}; |
239 | |
732152b1 |
240 | # Write the reading. |
241 | my $reading = { 'text' => $group_readings->{$gst}, |
242 | 'missing' => wit_stringify( $lacunose ), |
c4a4fb1b |
243 | 'group' => $gst }; # This will change if we find no conflict |
244 | if( $is_conflicted ) { |
245 | $reading->{'conflict'} = $conflict->{$group_readings->{$gst}} |
732152b1 |
246 | } else { |
c4a4fb1b |
247 | # Save the relevant subgraph. |
248 | $subgraph{$gst} = { 'graph' => $part, |
249 | 'root' => $group_root, |
250 | 'reachable' => $reachable }; |
732152b1 |
251 | } |
252 | push( @{$variant_row->{'readings'}}, $reading ); |
d71100ed |
253 | } |
c4a4fb1b |
254 | |
255 | # Now that we have gone through all the rows, check the hypothetical |
256 | # readings for conflict if we haven't found one yet. |
257 | if( keys %subgraph && !keys %$conflict ) { |
258 | my @resolve; |
259 | foreach ( keys %contig ) { |
260 | next unless ref $contig{$_}; |
261 | if( scalar @{$contig{$_}} > 1 ) { |
262 | push( @resolve, $_ ); |
263 | } else { |
264 | $contig{$_} = scalar @{$contig{$_}} ? $contig{$_}->[0] : ''; |
265 | } |
266 | } |
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 ) { |
286 | next if $v eq $h; |
287 | if( !$still_reachable{$v} |
288 | && ( $contig{$v} eq $gst |
289 | || ( exists $still_contig{$v} |
290 | && $still_contig{$v} eq $gst ) ) ) { |
291 | # We need $h. |
292 | if( exists $still_contig{$h} ) { |
293 | # Conflict! |
294 | $conflict->{$group_readings->{$gst}} = |
295 | $group_readings->{$still_contig{$h}}; |
296 | } else { |
297 | $still_contig{$h} = $gst; |
298 | } |
299 | last; |
300 | } # else we don't need $h in this group. |
301 | } |
302 | } |
303 | } |
304 | |
305 | # Now, assuming no conflict, we have some hypothetical vertices in |
306 | # $still_contig that are the "real" group memberships. Replace these |
307 | # in $contig. |
308 | unless ( keys %$conflict ) { |
309 | foreach my $v ( keys %contig ) { |
310 | next unless ref $contig{$v}; |
311 | $contig{$v} = $still_contig{$v}; |
312 | } |
313 | } |
314 | } |
315 | |
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{$_} } |
321 | keys %contig; |
322 | $rdg->{'group'} = wit_stringify( \@members ); |
323 | } |
324 | |
08e0fb85 |
325 | $variant_row->{'genealogical'} = !( keys %$conflict ); |
732152b1 |
326 | return $variant_row; |
d71100ed |
327 | } |
328 | |
329 | # Add the variant, subject to a.c. representation logic. |
330 | # This assumes that we will see the 'main' version before the a.c. version. |
331 | sub add_variant_wit { |
332 | my( $arr, $wit, $acstr ) = @_; |
333 | my $skip; |
334 | if( $wit =~ /^(.*)\Q$acstr\E$/ ) { |
335 | my $real = $1; |
336 | $skip = grep { $_ =~ /^\Q$real\E$/ } @$arr; |
337 | } |
338 | push( @$arr, $wit ) unless $skip; |
339 | } |
340 | |
341 | # Return an answer if the variant is useful, i.e. if there are at least 2 variants |
342 | # with at least 2 witnesses each. |
343 | sub useful_variant { |
344 | my( $readings ) = @_; |
345 | my $total = keys %$readings; |
346 | foreach my $var ( keys %$readings ) { |
347 | $total-- if @{$readings->{$var}} == 1; |
348 | } |
349 | return( undef, undef ) if $total <= 1; |
350 | my( $groups, $text ); |
351 | foreach my $var ( keys %$readings ) { |
352 | push( @$groups, $readings->{$var} ); |
353 | push( @$text, $var ); |
354 | } |
355 | return( $groups, $text ); |
356 | } |
357 | |
358 | # Take an array of witness groupings and produce a string like |
359 | # ['A','B'] / ['C','D','E'] / ['F'] |
360 | |
361 | sub wit_stringify { |
362 | my $groups = shift; |
363 | my @gst; |
364 | # If we were passed an array of witnesses instead of an array of |
365 | # groupings, then "group" the witnesses first. |
366 | unless( ref( $groups->[0] ) ) { |
367 | my $mkgrp = [ $groups ]; |
368 | $groups = $mkgrp; |
369 | } |
370 | foreach my $g ( @$groups ) { |
371 | push( @gst, '[' . join( ',', map { "'$_'" } @$g ) . ']' ); |
372 | } |
373 | return join( ' / ', @gst ); |
374 | } |
375 | |
376 | 1; |