pytrendy 1.2.0.dev5__tar.gz → 1.2.0.dev6__tar.gz

{pytrendy-1.2.0.dev5 → pytrendy-1.2.0.dev6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pytrendy
-Version: 1.2.0.dev5
+Version: 1.2.0.dev6
 Summary: Trend Detection in Python. Applicable for real-world industry use cases in time series.
 License: MIT License
          

{pytrendy-1.2.0.dev5 → pytrendy-1.2.0.dev6}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "pytrendy"
-version = "1.2.0.dev5"
+version = "1.2.0.dev6"
 description = "Trend Detection in Python. Applicable for real-world industry use cases in time series."
 authors = [
     { name = "Russell Sammut Bonnici", email = "r.sammutbonnici@gmail.com" },

{pytrendy-1.2.0.dev5 → pytrendy-1.2.0.dev6}/pytrendy/post_processing/segments_get.py

@@ -72,6 +72,8 @@ def get_segments(df: pd.DataFrame) -> list[dict]:
                     , 'end': end.strftime('%Y-%m-%d')
                 })
                 segment_length=0
+            else:
+                segment_length = 0  # Reset on fail: prevent length bleed into next segment
 
         direction_prev = direction
         segment_length_prev = segment_length

{pytrendy-1.2.0.dev5 → pytrendy-1.2.0.dev6}/pytrendy/post_processing/segments_refine/gradual_expand_contract.py

@@ -71,6 +71,33 @@ def expand_contract_segments(df: pd.DataFrame, value_col: str, segments: list[di
         else:
             continue
 
+        # Avoid orphaning a peak/trough that no adjacent segment covers.
+        # The +1 day pushes start past the extremum, assuming it belongs to
+        # the neighbour. When no neighbour is near and the extremum holds a
+        # significantly different value, the +1 day skips a genuine drop/rise.
+        # Skip this when the previous segment is Noise — noise boundaries are
+        # deliberately fuzzy and shouldn't anchor the orphan test.
+        prev_seg = segments_refined[i - 1] if i > 0 else None
+        prev_is_noise = prev_seg is not None and prev_seg.get('direction') == 'Noise'
+        if segment['direction'] in ('Up', 'Down') and i > 0:
+            prev_end = pd.to_datetime(segments_refined[i - 1]['end'])
+            extremum = pd.to_datetime(new_start) - pd.Timedelta(days=1)
+            distance = (extremum - prev_end).days
+            # Skip orphan check when previous segment is Noise AND the Noise
+            # is close (within 3 days) to the extremum — noise boundaries are
+            # deliberately fuzzy in that case.  When Noise is far away, the
+            # extremum is genuinely orphaned and should be captured.
+            # For non-Noise neighbours, keep the original distance > 1 gate.
+            if prev_is_noise and distance <= 3:
+                pass
+            elif (not prev_is_noise and distance > 1) or (prev_is_noise and distance > 3):
+                if extremum in df.index and new_start in df.index:
+                    extremum_val = df.loc[extremum, value_col]
+                    start_val = df.loc[new_start, value_col]
+                    max_abs = df[value_col].abs().max()
+                    if max_abs > 0 and abs(extremum_val - start_val) > 0.2 * max_abs:
+                        new_start -= pd.Timedelta(days=1)
+
         # Check for any inversions
         start_inverted = (new_start >= pd.to_datetime(segment['end']))
         end_inverted = (new_end <= pd.to_datetime(segment['start']))