fixed alignment_obsolete, rewrite alignment with new approach

evaluation/alignment.py

@@ -30,7 +30,7 @@ def procedure(anchor, subsec, S_arr, subidx):
 # Input: path1, path2
 # Output: aligned array of SRTsegment corresponding to path1 path2
 # Note: Modify comment with .source_text to get output array with string only
-def alignment(pred_path, gt_path):
+def alignment_obsolete(pred_path, gt_path):
     empt = SrtSegment([0,'00:00:00,000 --> 00:00:00,000','','',''])
     pred = SrtScript.parse_from_srt_file(pred_path).segments
     gt = SrtScript.parse_from_srt_file(gt_path).segments
@@ -67,12 +67,14 @@ def alignment(pred_path, gt_path):
                 gt_arr.append(empt)  # append filler
                 idx_t -= 1  # reset ground truth index
             else:
-                gt_arr.append(gs)#.source_text
+                
                 if gs.end >= ps.start:
+                    gt_arr.append(gs)#.source_text
                     pred_arr.append(ps)#.source_text
                     idx_p = procedure(gs, pred, pred_arr, idx_p + 1)
-                else:  # filler pairing
-                    pred_arr.append(empt)
+                else:
+                    gt_arr[len(gt_arr) - 1] += gs#.source_text
+                    #pred_arr.append(empt)
                     idx_p -= 1
         else:
             # same overlap checking procedure
@@ -81,19 +83,51 @@ def alignment(pred_path, gt_path):
                 pred_arr.append(empt)  # filler
                 idx_p -= 1  # reset
             else:
-                pred_arr.append(ps)#.source_text
                 if ps.end >= gs.start:
+                    pred_arr.append(ps)#.source_text
                     gt_arr.append(gs)#.source_text
                     idx_t = procedure(ps, gt, gt_arr, idx_t + 1)
                 else:  # filler pairing
-                    gt_arr.append(empt)
+                    pred_arr[len(pred_arr) - 1] += ps
                     idx_t -= 1
 
         idx_p += 1
         idx_t += 1
-    #for a in pred_arr:
-    #    print(a.source_text)
+    #for a in gt_arr:
+    #    print(a.translation)
+    return zip(pred_arr, gt_arr)
+
+# Input: path1, path2, threshold = 0.5 sec by default
+# Output: aligned array of SRTsegment corresponding to path1 path2
+def alignment(pred_path, gt_path, threshold=0.5):
+    empt = SrtSegment([0, '00:00:00,000 --> 00:00:00,000', '', '', ''])
+    pred = SrtScript.parse_from_srt_file(pred_path).segments
+    gt = SrtScript.parse_from_srt_file(gt_path).segments
+    pred_arr, gt_arr = [], []
+    idx_p, idx_t = 0, 0
+
+    while idx_p < len(pred) or idx_t < len(gt):
+        ps = pred[idx_p] if idx_p < len(pred) else empt
+        gs = gt[idx_t] if idx_t < len(gt) else empt
+
+        # Merging sequence for pred
+        while idx_p + 1 < len(pred) and abs(pred[idx_p + 1].end - gs.end) <= threshold:
+            ps += pred[idx_p + 1]
+            idx_p += 1
+
+        # Merging sequence for gt
+        while idx_t + 1 < len(gt) and abs(gt[idx_t + 1].end - ps.end) <= threshold:
+            gs += gt[idx_t + 1]
+            idx_t += 1
+
+        # Append to the result arrays
+        pred_arr.append(ps)
+        gt_arr.append(gs)
+        idx_p += 1
+        idx_t += 1
+
     return zip(pred_arr, gt_arr)
 
+
 #  Test Case
 #alignment('test_translation_zh.srt', 'test_translation_bi.srt')