batchalign 0.8.2.post3__tar.gz → 0.8.2.post6__tar.gz

{batchalign-0.8.2.post3/batchalign.egg-info → batchalign-0.8.2.post6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.8.2.post3
+Version: 0.8.2.post6
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

{batchalign-0.8.2.post3 → batchalign-0.8.2.post6}/batchalign/cli/dispatch.py

@@ -220,22 +220,19 @@ def _run_pipeline_for_file(command, pipeline, file, output, loader_info, writer_
         main_doc = CHATFile(path=str(p)).doc
         gold_doc = CHATFile(path=str(gold_path), special_mor_=True).doc
 
-        # Pipeline: morphosyntax(main) -> compare -> compare_analysis
+        # Pipeline: compare (runs morphosyntax internally) -> compare_analysis
         result = pipeline(main_doc, callback=progress_callback, gold=gold_doc)
 
         # Write annotated CHAT
         CHATFile(doc=result["doc"]).write(output,
                                           merge_abbrev=local_kwargs.get("merge_abbrev", False))
 
-        # Write metrics CSV
-        import csv
+        # Write metrics as JSON sidecar for consolidated CSV later
+        import json as _json
         metrics = result["metrics"]
-        csv_path = P(output).with_suffix(".compare.csv")
-        with open(csv_path, 'w', newline='') as f:
-            writer = csv.writer(f)
-            writer.writerow(["metric", "value"])
-            for k, v in metrics.items():
-                writer.writerow([k, v])
+        json_path = P(output).with_suffix(".compare.json")
+        with open(json_path, 'w') as f:
+            _json.dump(metrics, f)
 
     elif command == "opensmile":
         from batchalign.document import Document
@@ -388,7 +385,7 @@ Cmd2Task = {
     "coref": "coref",
     "translate": "translate",
     "opensmile": "opensmile",
-    "compare": "morphosyntax,compare,compare_analysis",
+    "compare": "compare,compare_analysis",
 }
 
 # this is the main runner used by all functions
@@ -1049,6 +1046,38 @@ def _dispatch(command, lang, num_speakers,
         if not pool_mode_enabled:
             _persist_memory_history()
 
+    # --- Consolidate per-file compare metrics into a single CSV ---
+    if command == "compare" and outputs:
+        import csv
+        import json as _json
+
+        all_metrics = []  # list of (filename, metrics_dict)
+        all_keys = []     # ordered superset of metric keys
+
+        for output_path in outputs:
+            json_path = Path(output_path).with_suffix(".compare.json")
+            if not json_path.exists():
+                continue
+            try:
+                with open(json_path, 'r') as f:
+                    metrics = _json.load(f)
+                rel = os.path.relpath(output_path, out_dir)
+                all_metrics.append((rel, metrics))
+                for k in metrics:
+                    if k not in all_keys:
+                        all_keys.append(k)
+                json_path.unlink()
+            except Exception:
+                pass
+
+        if all_metrics:
+            csv_path = Path(out_dir) / "compare.csv"
+            with open(csv_path, 'w', newline='') as f:
+                writer = csv.writer(f)
+                writer.writerow(["file"] + all_keys)
+                for filename, metrics in all_metrics:
+                    writer.writerow([filename] + [metrics.get(k, "") for k in all_keys])
+
     if len(errors) > 0:
         C.print()
         for file, trcbk, err, captured in errors:

{batchalign-0.8.2.post3 → batchalign-0.8.2.post6}/batchalign/pipelines/analysis/compare.py

@@ -4,7 +4,7 @@ Engines for transcript comparison against gold-standard references.
 
 CompareEngine (PROCESSING): Aligns main vs gold transcripts word-by-word
 using the same conform/match_fn logic as WER evaluation, then annotates
-each main utterance with comparison tokens (%xsrep / %xsmor).
+each gold utterance with comparison tokens (%xsrep / %xsmor).
 
 CompareAnalysisEngine (ANALYSIS): Reads the comparison annotations and
 computes error-rate metrics for CSV output.
@@ -208,6 +208,11 @@ def _find_best_segment(gold_tokens, main_tokens, mfn):
                 elif len_delta == best_len_delta and end > best[1]:
                     best = (start, end)
 
+    # If no tokens overlap at all, return an empty window so the caller
+    # doesn't consume main tokens that belong to a later gold utterance.
+    if best_score <= 0:
+        return 0, 0
+
     return best
 
 
@@ -249,6 +254,15 @@ def conform_with_mapping(words, conform_fn):
 class CompareEngine(BatchalignEngine):
     tasks = [Task.COMPARE]
 
+    def __init__(self):
+        self._stanza = None
+
+    def _get_stanza(self):
+        if self._stanza is None:
+            from batchalign.pipelines.morphosyntax import StanzaEngine
+            self._stanza = StanzaEngine()
+        return self._stanza
+
     def process(self, doc, **kwargs):
         gold = kwargs.get("gold")
         if not gold or not isinstance(gold, Document):
@@ -256,6 +270,12 @@ class CompareEngine(BatchalignEngine):
                 f"CompareEngine requires a 'gold' Document kwarg, got '{type(gold)}'"
             )
 
+        # --- 0. Run morphosyntax on both docs ---
+        stanza = self._get_stanza()
+        non_gold_kwargs = {k: v for k, v in kwargs.items() if k != "gold"}
+        doc = stanza.process(doc, **non_gold_kwargs)
+        gold = stanza.process(gold, **non_gold_kwargs)
+
         # --- 1. Extract words from main utterances ---
         main_utterances = [
             u for u in doc.content if isinstance(u, Utterance)
@@ -343,16 +363,14 @@ class CompareEngine(BatchalignEngine):
                     gold_form = gold_info[orig_gold_idx][2]
                     last_gold_form_idx = gold_form_idx
 
+                    # Copy timing from main; keep gold's own morphology
                     if main_form.time is not None:
                         gold_form.time = main_form.time
-                    if main_form.morphology is not None:
-                        gold_form.morphology = main_form.morphology
-                    if main_form.dependency is not None:
-                        gold_form.dependency = main_form.dependency
+                    gold_form.dependency = None
 
                     utt_positioned[utt_idx].append((gold_form_idx, CompareToken(
                         text=item.key,
-                        pos=_get_pos(main_form),
+                        pos=_get_pos(gold_form),
                         status="match"
                     )))
                     local_main_cursor += 1
@@ -365,6 +383,8 @@ class CompareEngine(BatchalignEngine):
                         gold_form = gold_info[orig_gold_idx][2]
                         last_gold_form_idx = gold_form_idx
 
+                        gold_form.dependency = None
+
                         utt_positioned[utt_idx].append((gold_form_idx, CompareToken(
                             text=item.key,
                             pos=_get_pos(gold_form),
@@ -401,6 +421,7 @@ class CompareEngine(BatchalignEngine):
                     form.morphology = [Morphology(
                         pos="PUNCT", lemma=form.text.strip(), feats=""
                     )]
+                form.dependency = None
             utt_positioned[utt_idx].sort(key=lambda x: x[0])
 
         # --- 7. Set comparison on each gold utterance ---

{batchalign-0.8.2.post3 → batchalign-0.8.2.post6}/batchalign/version

@@ -1,3 +1,3 @@
-0.8.2-post.3
+0.8.2-post.6
 March 28 2026
 fix bugs

{batchalign-0.8.2.post3 → batchalign-0.8.2.post6/batchalign.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.8.2.post3
+Version: 0.8.2.post6
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu