Performance increases

Temporarily switch to the c-implementation of python-levenshtein for
editops calculatation. Also added some variables, caching and type
changes for performance gains.

qurator/dinglehopper/flexible_character_accuracy.py

@@ -17,9 +17,10 @@ from functools import lru_cache, reduce
 from itertools import product, takewhile
 from typing import List, Tuple, Optional
 
+from Levenshtein import editops
 from multimethod import multimethod
 
-from . import editops, ExtractedText
+from . import ExtractedText
 
 if sys.version_info.minor == 5:
     from .flexible_character_accuracy_ds_35 import (
@@ -170,10 +171,21 @@ def match_gt_line(
     """
     min_penalty = float("inf")
     best_match, best_ocr = None, None
+    gt_line_length = gt_line.length
+    gt_line_start = gt_line.start
     for ocr_line in ocr_lines:
         match = match_lines(gt_line, ocr_line)
         if match:
-            penalty = calculate_penalty(gt_line, ocr_line, match, coef)
+            penalty = calculate_penalty(
+                gt_line_length,
+                ocr_line.length,
+                gt_line_start,
+                ocr_line.start,
+                match.gt.start,
+                match.ocr.start,
+                match.dist,
+                coef,
+            )
             if penalty < min_penalty:
                 min_penalty, best_match, best_ocr = penalty, match, ocr_line
     return best_match, best_ocr
@@ -234,7 +246,7 @@ def match_lines(gt_line: "Part", ocr_line: "Part") -> Optional[Match]:
     for i, gt_part in gt_parts:
         for j, ocr_part in ocr_parts:
             match = distance(gt_part, ocr_part)
-            edit_dist = score_edit_distance(match)
+            edit_dist = score_edit_distance(match.dist)
             if edit_dist < min_edit_dist and match.dist.replace < min_length:
                 min_edit_dist = edit_dist
                 best_match = match
@@ -248,13 +260,13 @@ def match_lines(gt_line: "Part", ocr_line: "Part") -> Optional[Match]:
                 gt_line.substring(rel_start=best_i, rel_end=best_i + k),
                 ocr_line.substring(rel_start=best_j, rel_end=best_j + k),
             )
-            edit_dist = score_edit_distance(match)
+            edit_dist = score_edit_distance(match.dist)
             if edit_dist < min_edit_dist and match.dist.replace < min_length:
                 min_edit_dist = edit_dist
                 best_match = match
     # is delete a better option?
     match = distance(gt_line, Part(text="", line=ocr_line.line, start=ocr_line.start))
-    edit_dist = score_edit_distance(match)
+    edit_dist = score_edit_distance(match.dist)
     if edit_dist < min_edit_dist:
         best_match = match
 
@@ -278,18 +290,26 @@ def distance(gt: "Part", ocr: "Part") -> Match:
     return Match(gt=gt, ocr=ocr, dist=Distance(**edits), ops=ops)
 
 
-def score_edit_distance(match: Match) -> int:
+def score_edit_distance(dist: Distance) -> int:
     """Calculate edit distance for a match.
 
     Formula: $deletes + inserts + 2 * replacements$
 
     :return: Sum of deletes, inserts and replacements.
     """
-    return match.dist.delete + match.dist.insert + 2 * match.dist.replace
+    return dist.delete + dist.insert + 2 * dist.replace
 
 
+@lru_cache(1000000)
 def calculate_penalty(
-    gt: "Part", ocr: "Part", match: Match, coef: Coefficients
+    gt_length: int,
+    ocr_length: int,
+    gt_start: int,
+    ocr_start: int,
+    gt_match_start: int,
+    ocr_match_start: int,
+    dist: Distance,
+    coef: Coefficients,
 ) -> float:
     """Calculate the penalty for a given match.
 
@@ -297,12 +317,12 @@ def calculate_penalty(
 
     :return: Penalty for the given match.
     """
-    min_edit_dist = score_edit_distance(match)
+    min_edit_dist = score_edit_distance(dist)
-    length_diff = abs(gt.length - ocr.length)
+    length_diff = abs(gt_length - ocr_length)
-    substring_length = min(gt.length, ocr.length)
+    substring_length = min(gt_length, ocr_length)
     offset = 0.0
     if length_diff > 1:
-        substring_pos = max(match.gt.start - gt.start, match.ocr.start - ocr.start)
+        substring_pos = max(gt_match_start - gt_start, ocr_match_start - ocr_start)
         offset = length_diff / 2 - abs(substring_pos - length_diff / 2)
     return (
         min_edit_dist * coef.edit_dist
@@ -428,4 +448,4 @@ class Part(PartVersionSpecific):
         """
         text = self.text[rel_start:rel_end]
         start = self.start + rel_start
-        return Part(**{**self._asdict(), "text": text, "start": start})
+        return Part(line=self.line, text=text, start=start)

qurator/dinglehopper/tests/test_flexible_character_accuracy.py

@@ -70,9 +70,9 @@ SIMPLE_EDITS = [
     (Part(text="a"), Part(text="a"), Distance(match=1)),
     (Part(text="aaa"), Part(text="aaa"), Distance(match=3)),
     (
-        Part(text="abcd"),
+        Part(text="abbbbcd"),
-        Part(text="beed"),
+        Part(text="bbbbede"),
-        Distance(match=2, replace=1, insert=1, delete=1),
+        Distance(match=5, replace=1, insert=1, delete=1),
     ),
 ]
 

qurator/dinglehopper/tests/test_integ_flexible_character_accuracy.py

@@ -50,7 +50,6 @@ def test_reading_order_settings(file, expected_text):
         assert ocr == expected_text
 
 
-@pytest.mark.skip(reason="Need to check performance first.")
 @pytest.mark.integration
 @pytest.mark.parametrize(
     "gt,ocr,expected",

requirements.txt

@@ -9,3 +9,4 @@ ocrd >= 2.20.1
 attrs
 multimethod == 1.3  # latest version to officially support Python 3.5
 tqdm
+python-levenshtein