Update src/streamlit_app.py

src/streamlit_app.py

@@ -104,7 +104,7 @@ SIZE_COLORS = {
 def fetch_whale_data(sym: str) -> pd.DataFrame:
     try:
         ticker = yf.Ticker(sym)
-        # 1-minute granularity preserved for accurate size-tier classification;
+        # 1-minute granularity for accurate size-tier classification;
         # Yahoo only serves 1m bars for the trailing 7 days.
         return ticker.history(period="7d", interval="1m")
     except Exception:
@@ -150,14 +150,15 @@ def render_whale_monitor():
 def render_whale_single_ticker():
     st.title("🐋 Whale Monitor — Single Ticker")
     st.caption(
-        "Flags unusual 1-minute volume on the public exchange tape over the trailing 7 days. "
+        "Flags unusual 1-minute volume on the public exchange tape over the trailing 7 days, "
+        "compared against each day's own volume baseline. "
         "This is a lit-market 'unusual volume' heuristic, not off-exchange dark pool print data."
     )
 
     with st.sidebar:
         st.subheader("Single Ticker Settings")
         symbol = st.text_input("Ticker", value="AIIO", key="wm_ticker").strip().upper()
-        multiplier = st.slider("Sensitivity (x median volume)", 1.0, 10.0, 2.0, 0.5, key="wm_mult")
+        multiplier = st.slider("Sensitivity (x daily median volume)", 1.0, 10.0, 2.0, 0.5, key="wm_mult")
         whale_threshold = st.number_input("Whale alert threshold (volume)", value=50000, step=5000, key="wm_thresh")
         show_all = st.checkbox("Show all candles (ignore threshold)", value=False, key="wm_all")
         exclude_open = st.checkbox("Exclude opening auction print (each day's first bar)", value=True, key="wm_exopen")
@@ -174,13 +175,27 @@ def render_whale_single_ticker():
         return
 
     df = df.copy()
-    first_day = df.index.normalize().min()
-    last_day = df.index.normalize().max()
+
+    # Compute the day-grouping column once, up front, and keep it around
+    # for both the "exclude opening bar" step and the daily-median calc
+    # below. Only drop it right at the end (with errors="ignore" as a
+    # safety net in case a pandas version strips it during groupby.apply).
+    df["_day"] = df.index.normalize()
+
+    first_day = df["_day"].min()
+    last_day = df["_day"].max()
 
     if exclude_open:
-        df["_day"] = df.index.normalize()
-        df = df.groupby("_day", group_keys=False).apply(lambda g: g.iloc[1:] if len(g) > 1 else g)
-        df = df.drop(columns="_day")
+        def _drop_first_bar(g):
+            return g.iloc[1:] if len(g) > 1 else g
+
+        df = df.groupby("_day", group_keys=False).apply(_drop_first_bar, include_groups=True)
+
+        # Some pandas versions strip the grouping column from the result
+        # of groupby(...).apply(...) even with include_groups=True on
+        # certain releases; rebuild it defensively if that happened.
+        if "_day" not in df.columns:
+            df["_day"] = df.index.normalize()
 
     if df.empty:
         st.error("No data for the selected 7-day window.")
@@ -191,28 +206,45 @@ def render_whale_single_ticker():
         st.error("Not enough volume data after filtering.")
         return
 
-    median_vol = df[df["Volume"] > 0]["Volume"].median()
-    mean_vol   = df[df["Volume"] > 0]["Volume"].mean()
+    # --- Per-day median/mean, not pooled across the whole week ---
+    # This keeps "WHALE"/"INSTITUTIONAL"/etc. classification accurate to
+    # each day's own volume regime, rather than diluting a high-volume day
+    # into a quiet week's baseline (or vice versa).
+    vol_positive = df["Volume"] > 0
+    daily_median = df[vol_positive].groupby(df["_day"])["Volume"].median()
+    daily_mean   = df[vol_positive].groupby(df["_day"])["Volume"].mean()
+
+    df["Daily_Median_Vol"] = df["_day"].map(daily_median)
+    df["Daily_Mean_Vol"]   = df["_day"].map(daily_mean)
 
     df["Price_Change"] = df["Close"].diff()
     df["Pct_Change"]   = df["Close"].pct_change() * 100
-    df["Vol_Ratio"]    = df["Volume"] / median_vol
-    df["is_notable"]   = df["Volume"] > (median_vol * multiplier) if not show_all else True
+    df["Vol_Ratio"]    = df["Volume"] / df["Daily_Median_Vol"]
+    df["is_notable"]   = df["Volume"] > (df["Daily_Median_Vol"] * multiplier) if not show_all else True
 
     df[["Delta", "Sentiment"]] = df.apply(lambda r: pd.Series(get_sentiment(r)), axis=1)
     df["Cum_Delta"] = df["Delta"].cumsum()
-    df["Size"]      = df["Volume"].apply(lambda v: classify_size(v, median_vol))
+    df["Size"] = df.apply(lambda r: classify_size(r["Volume"], r["Daily_Median_Vol"]), axis=1)
+
+    # Overall (pooled) figures shown in the top metrics — labeled clearly
+    # as an average-of-days summary, not the classification baseline.
+    median_vol = daily_median.mean()
+    mean_vol   = daily_mean.mean()
 
     st.subheader(f"{symbol} — Last 7 Days: {first_day.date()} to {last_day.date()}")
     c1, c2, c3, c4 = st.columns(4)
-    c1.metric("Median 1-min Volume", f"{median_vol:,.0f}")
-    c2.metric("Mean 1-min Volume",   f"{mean_vol:,.0f}")
-    c3.metric("Flag Threshold",      f"{median_vol * multiplier:,.0f}")
+    c1.metric("Avg of Daily Median Vol", f"{median_vol:,.0f}")
+    c2.metric("Avg of Daily Mean Vol",   f"{mean_vol:,.0f}")
+    c3.metric("Flag Threshold (avg)",    f"{median_vol * multiplier:,.0f}")
     c4.metric(
         "Last Price",
         f"${df['Close'].iloc[-1]:.2f}",
         f"{df['Pct_Change'].iloc[-1]:+.2f}%" if pd.notna(df["Pct_Change"].iloc[-1]) else None,
     )
+    st.caption(
+        "Size-tier classification (WHALE/INSTITUTIONAL/etc.) and the Notable Activity table below use "
+        "each bar's **own day's** median volume, not the pooled 7-day figure shown above."
+    )
 
     st.markdown("### Notable Activity")
     notable = df[df["is_notable"]].copy()
@@ -248,7 +280,7 @@ def render_whale_single_ticker():
             name=size.title(),
             line=dict(color=color, width=2),
         ))
-    # optional: thin dotted total line for reference
+    # thin dotted total line for reference
     fig.add_trace(go.Scatter(
         x=df.index,
         y=df["Cum_Delta"],
@@ -307,7 +339,8 @@ def render_whale_single_ticker():
     st.dataframe(pd.DataFrame(rows), use_container_width=True, hide_index=True)
 
     st.caption(
-        "Data source: yfinance (Yahoo Finance), unofficial and delayed. 1-minute bars, 7-day window. "
+        "Data source: yfinance (Yahoo Finance), unofficial and delayed. 1-minute bars, 7-day window, "
+        "classified against each day's own volume baseline. "
         "Cached for 2 minutes — use the Refresh button in the sidebar for the latest data."
     )