batchalign 0.8.2.post4__tar.gz → 0.8.2.post7__tar.gz

{batchalign-0.8.2.post4/batchalign.egg-info → batchalign-0.8.2.post7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.8.2.post4
+Version: 0.8.2.post7
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

{batchalign-0.8.2.post4 → batchalign-0.8.2.post7}/batchalign/cli/dispatch.py

@@ -227,15 +227,12 @@ def _run_pipeline_for_file(command, pipeline, file, output, loader_info, writer_
         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
@@ -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.post4 → batchalign-0.8.2.post7}/batchalign/pipelines/analysis/compare.py

@@ -329,6 +329,10 @@ class CompareEngine(BatchalignEngine):
         # window in the remaining main tokens, then run Levenshtein inside
         # that window to produce the annotations.
         utt_positioned = {i: [] for i in range(len(gold_utterances))}
+        # Track which main (child) forms land in each gold utterance's window
+        utt_main_forms = {i: [] for i in range(len(gold_utterances))}
+        # Track which main utterance(s) contribute to each gold utterance
+        utt_main_speakers = {i: [] for i in range(len(gold_utterances))}
         search_start = 0
 
         for utt_idx in range(len(gold_utterances)):
@@ -345,6 +349,17 @@ class CompareEngine(BatchalignEngine):
             abs_start = search_start + win_start
             abs_end = search_start + win_end
 
+            # Collect unique main forms in the window (deduplicate
+            # across conformed expansions that map to the same form)
+            seen_main = set()
+            for j in range(abs_start, abs_end):
+                orig_idx = main_map[j]
+                if orig_idx not in seen_main:
+                    seen_main.add(orig_idx)
+                    m_utt_idx, m_form_idx, m_form = main_info[orig_idx]
+                    utt_main_forms[utt_idx].append(m_form)
+                    utt_main_speakers[utt_idx].append(m_utt_idx)
+
             # Align the chosen window against this gold utterance
             window_main = conformed_main[abs_start:abs_end]
             utt_alignment = align(window_main, g_tokens, False, match_fn)
@@ -363,11 +378,6 @@ 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
-                    gold_form.dependency = None
-
                     utt_positioned[utt_idx].append((gold_form_idx, CompareToken(
                         text=item.key,
                         pos=_get_pos(gold_form),
@@ -383,8 +393,6 @@ 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),
@@ -415,21 +423,41 @@ class CompareEngine(BatchalignEngine):
                     pos="PUNCT",
                     status="match"
                 )))
-                # Ensure punctuation forms have morphology so %mor includes
-                # the final delimiter (matches generator's PUNCT|<char> logic)
-                if form.morphology is None:
-                    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 ---
+        # --- 7. Replace each gold utterance's main line with the child's
+        #         (main/proposal) forms and set the child's speaker. ---
         for utt_idx, utt in enumerate(gold_utterances):
             tokens = [tok for _, tok in utt_positioned[utt_idx]]
             utt.comparison = tokens if tokens else None
 
-            timed_forms = [form for form in utt.content if form.time is not None]
+            # Replace content with the child/main forms from the window
+            child_forms = utt_main_forms[utt_idx]
+            if child_forms:
+                # Grab the ending punctuation from gold so the utterance
+                # delimiter is preserved
+                gold_ending = [
+                    f for f in utt.content
+                    if f.text.strip() in ENDING_PUNCT
+                ]
+                # Strip dependency from child forms (morphosyntax was run
+                # on the main doc separately; gold structure differs)
+                for f in child_forms:
+                    f.dependency = None
+                utt.content = child_forms + gold_ending
+
+                # Set speaker from the child utterance that contributed
+                # the most forms to this window
+                speaker_indices = utt_main_speakers[utt_idx]
+                if speaker_indices:
+                    # Pick the most common main utterance index
+                    majority_utt_idx = max(
+                        set(speaker_indices), key=speaker_indices.count
+                    )
+                    utt.tier = main_utterances[majority_utt_idx].tier
+
+            # Derive utterance timing from the (now-child) forms
+            timed_forms = [f for f in utt.content if f.time is not None]
             if timed_forms:
                 utt.time = (timed_forms[0].time[0], timed_forms[-1].time[1])
                 utt.text = None

batchalign-0.8.2.post7/batchalign/version

@@ -0,0 +1,3 @@
+0.8.2-post.7
+April 01 2026
+more compare updates

{batchalign-0.8.2.post4 → batchalign-0.8.2.post7/batchalign.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.8.2.post4
+Version: 0.8.2.post7
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

batchalign-0.8.2.post4/batchalign/version

@@ -1,3 +0,0 @@
-0.8.2-post.4
-March 30 2026
-More compare bug fixes