Merge pull request #131 from tmbdev/levenshtein

[econf] Handle possibility that filtered GT is empty

ocrolib/edist.py

@@ -21,6 +21,8 @@ def xlevenshtein(a,b,context=1):
     """Calculates the Levensthein distance between a and b
     and generates a list of differences by context."""
     n, m = len(a), len(b)
+    assert m>0 # xlevenshtein should only be called with non-empty b string (ground truth)
+    if a == b: return 0,[] # speed up for the easy case
     sources = empty((m+1,n+1),object)
     sources[:,:] = None
     dists = 99999*ones((m+1,n+1))

ocropus-econf

@@ -48,13 +48,19 @@ def process1(fname):
     else:
         missing = len(gt)
         txt = ""
-    if args.confusion>0 or args.allconf is not None:
+    # Also the ground truth cannot be empty, it is possible that
+    # after filtering (args.kind) the gt string is empty.
+    if len(gt) == 0:
+        err = len(txt)
+        if(len(txt)>0):
+            cs = [(txt,'_'*len(txt))]
+        else:
+            cs = []
+    else:
         err,cs = edist.xlevenshtein(txt,gt,context=args.context)
+    if args.confusion>0 or args.allconf is not None:
         for u,v in cs:
             counts[(u,v)] += 1
-    else:
-        err = edist.levenshtein(txt,gt)
-    #assert err==xerr
     return fname,err,len(gt),missing,counts
 
 outputs = ocrolib.parallel_map(process1,args.files,parallel=args.parallel,chunksize=10)
@@ -89,11 +95,15 @@ if allconf is not None: allconf.close()
 sys.stderr.write("errors    %8d\n"%errs)
 sys.stderr.write("missing   %8d\n"%missing)
 sys.stderr.write("total     %8d\n"%total)
-sys.stderr.write("err       %8.3f %%\n"%(errs*100.0/total))
-sys.stderr.write("errnomiss %8.3f %%\n"%((errs-missing)*100.0/total))
+if (total>0):
+    sys.stderr.write("err       %8.3f %%\n"%(errs*100.0/total))
+    sys.stderr.write("errnomiss %8.3f %%\n"%((errs-missing)*100.0/total))
 
 if args.confusion>0:
     for (a,b),v in counts.most_common(args.confusion):
         print("%d\t%s\t%s" % (v, a, b))
 
-print(errs * 1.0 / total)
+if (total>0):
+    print(errs * 1.0 / total)
+else:
+    print("Nothing to compare")

tests/run-unit

@@ -0,0 +1,13 @@
+#!/usr/bin/env python
+
+from ocrolib import edist
+def testLevenshtein(a, b, should):
+    if edist.levenshtein(a, b) == should:
+        print 'ok - levenshtein(%s, %s) == %s' % (a,b,should)
+    else:
+        print 'not ok - levenshtein(%s, %s) == %s' % (a,b,should)
+testLevenshtein('a', 'a', 0)
+testLevenshtein('', '', 0)
+testLevenshtein('a', '', 1)
+testLevenshtein('', 'a', 1)
+testLevenshtein('aa', 'aaaaaa', 4)