Merge pull request 'fix(site-finder): volume-weighted median_price_m2 (#1511) + consistent-period MTS (#1512)' (#1690) from fix/match-economics-1511-1512 into main

site-finder/06_match_economics.py

@@ -5,7 +5,10 @@ Output tables:
   district_economics   — per Objective-район aggregates over last 90 days
   site_district        — which Objective-район each site belongs to
 """
-import sqlite3, pathlib, re, math
+import math
+import pathlib
+import re
+import sqlite3
 from difflib import SequenceMatcher
 
 DB = pathlib.Path(__file__).parent / "analysis.db"
@@ -43,7 +46,8 @@ CREATE TABLE IF NOT EXISTS site_district (
 """
 
 def normalize(s):
-    if not s: return ""
+    if not s:
+        return ""
     s = s.lower().strip()
     s = re.sub(r'^(жк|жилой\s+комплекс|жилой\s+квартал|жилые\s+кварталы)\s+', '', s)
     s = re.sub(r'["«»\'`]+', '', s)
@@ -53,11 +57,13 @@ def normalize(s):
 def fuzzy(a, b):
     return SequenceMatcher(None, normalize(a), normalize(b)).ratio()
 
-def hav(la1,lo1,la2,lo2):
+def hav(la1, lo1, la2, lo2):
-    R=6371000; p1,p2=math.radians(la1),math.radians(la2)
+    R = 6371000
-    dp=math.radians(la2-la1); dl=math.radians(lo2-lo1)
+    p1, p2 = math.radians(la1), math.radians(la2)
-    a=math.sin(dp/2)**2+math.cos(p1)*math.cos(p2)*math.sin(dl/2)**2
+    dp = math.radians(la2 - la1)
-    return 2*R*math.asin(math.sqrt(a))
+    dl = math.radians(lo2 - lo1)
+    a = math.sin(dp / 2) ** 2 + math.cos(p1) * math.cos(p2) * math.sin(dl / 2) ** 2
+    return 2 * R * math.asin(math.sqrt(a))
 
 def main():
     conn = sqlite3.connect(DB)
@@ -74,7 +80,8 @@ def main():
         best = (0.0, None)
         for proj in objective_projects:
             s = fuzzy(name, proj)
-            if s > best[0]: best = (s, proj)
+            if s > best[0]:
+                best = (s, proj)
         if best[0] >= 0.80:
             conn.execute("INSERT INTO jk_objective_match VALUES (?,?,?,?)",
                          (site_id, best[1], best[0], "fuzzy"))
@@ -90,17 +97,17 @@ def main():
     conn.execute("DELETE FROM district_economics")
     placeholders = ",".join(["?"]*len(last3))
 
-    # Latest stock per (project, corpus) — taken from the latest month each corpus appears
+    # Latest stock per (project, corpus) — taken from the single most-recent month in the window
+    # (#1512): using a per-corpus latest caused mixed periods (stale stock for inactive corpuses).
+    # Fix: pin stock to last3[0] (the most recent month) so numerator and denominator of MTS
+    # are aligned to the same period for all corpuses in the district.
+    latest_month = last3[0]
     latest_stock = {}
-    for r in conn.execute(f"""
+    for r in conn.execute("""
-        WITH ranked AS (
+        SELECT project, corpus, district, stock_m2, stock_lots
-            SELECT project, corpus, district, month, stock_m2, stock_lots,
+          FROM objective_corp_month
-                   ROW_NUMBER() OVER (PARTITION BY project, corpus ORDER BY month DESC) rn
+         WHERE month = ?
-              FROM objective_corp_month
+    """, (latest_month,)).fetchall():
-             WHERE month IN ({placeholders})
-        )
-        SELECT project, corpus, district, stock_m2, stock_lots FROM ranked WHERE rn=1
-    """, last3).fetchall():
         latest_stock[(r[0], r[1])] = (r[2], r[3] or 0, r[4] or 0)
 
     # Per-district aggregates from window
@@ -114,7 +121,8 @@ def main():
     by_dist = {}
     for r in rows:
         d = r[0]
-        if not d: continue
+        if not d:
+            continue
         by_dist.setdefault(d, []).append(r)
 
     for d, rs in by_dist.items():
@@ -124,9 +132,24 @@ def main():
         wn = sum((r[6] or 0) * (r[5] or 0) for r in rs)
         wd = sum((r[5] or 0) for r in rs)
         wp = wn / wd if wd else None
-        # median price (priced deals only)
+        # median price — volume-weighted by sold_volume_m2 (#1511)
-        prices = sorted([r[6] for r in rs if r[6] and (r[4] or 0) > 0])
+        # Each corpus×month row contributes avg_price_m2 weighted by sold_volume_m2 so that
-        med = prices[len(prices)//2] if prices else None
+        # high-volume months/rooms count proportionally (not equal-weight per row).
+        # Falls back to weight=1 for rows with priced deals but missing volume figure.
+        pw = [(r[6], max(r[5] or 0, 1.0)) for r in rs if r[6] and (r[4] or 0) > 0]
+        if pw:
+            pw.sort(key=lambda x: x[0])
+            total_w = sum(w for _, w in pw)
+            half = total_w / 2.0
+            cum = 0.0
+            med = None
+            for price, w in pw:
+                cum += w
+                if cum >= half:
+                    med = price
+                    break
+        else:
+            med = None
         # offer prices
         offers = [r[8] for r in rs if r[8] and r[8] > 0]
         avg_off = sum(offers)/len(offers) if offers else None
@@ -175,7 +198,8 @@ def main():
             # vote among k=5 nearest matched ЖК within 2 km
             cands = []
             for sid2, (lat2, lon2, dist2) in matched_idx.items():
-                if sid2 == sid: continue
+                if sid2 == sid:
+                    continue
                 d = hav(lat, lon, lat2, lon2)
                 cands.append((d, sid2, dist2))
             cands.sort()
@@ -187,8 +211,10 @@ def main():
                 rep = next(c for c in top if c[2] == top_dist)
                 obj_id = None
                 if rep[1] and rep[1].startswith("jk:"):
-                    try: obj_id = int(rep[1][3:])
+                    try:
-                    except: pass
+                        obj_id = int(rep[1][3:])
+                    except ValueError:
+                        pass
                 conn.execute("INSERT INTO site_district VALUES (?,?,?,?,?)",
                              (sid, top_dist, "knn_vote_5", obj_id, rep[0]))
     conn.commit()