fix(portfolio): handle NaN trade plot markers

tests/test_plotting.py

@@ -22,6 +22,7 @@
 
 import vectorbt as vbt
 from vectorbt.generic.plotting import Scatter, Bar, Histogram, Heatmap, Gauge, Box, Volume, TraceUpdater
+from vectorbt.portfolio.base import Portfolio
 from vectorbt.utils.figure import make_figure, make_subplots, Figure, FigureWidget
 
 
@@ -625,6 +626,45 @@ def test_trades_plot_pnl_pct_scale(self):
         # pct_scale=False: y-values are absolute PnL
         np.testing.assert_array_almost_equal(fig_abs.data[0].y, [1.78, 2.77])
 
+    def test_trades_plot_pnl_non_finite_returns(self):
+        """Non-finite returns should not invalidate Plotly marker metadata."""
+        # Cover the issue's exact case where no finite return is available for scaling.
+        open_pf = Portfolio.from_orders(
+            pd.Series([1.0, np.nan]),
+            pd.Series([1.0, 0.0]),
+            ffill_val_price=False,
+        )
+
+        # Exercise both the direct records API and its Portfolio convenience delegate.
+        for fig in (open_pf.get_trades().plot_pnl(), open_pf.plot_trade_pnl()):
+            assert isinstance(fig, BaseFigure)
+            nt = named_traces(fig)
+
+            # Without a finite reference, the open marker should use both lower bounds.
+            np.testing.assert_array_equal(nt["Open"].marker.size, [7.0])
+            np.testing.assert_array_equal(nt["Open"].marker.opacity, [0.75])
+
+        # One unknown open trade must not alter the scaling of finite closed trades.
+        close = pd.Series([1.0, 2.0, 1.0, 3.0, 1.0, np.nan])
+        size = pd.Series([1.0, -1.0, 1.0, -1.0, 1.0, 0.0])
+        pf = Portfolio.from_orders(close, size, ffill_val_price=False)
+
+        # Require identical marker behavior from the direct and delegated plotting paths.
+        for fig in (pf.get_trades().plot_pnl(), pf.plot_trade_pnl()):
+            assert isinstance(fig, BaseFigure)
+            nt = named_traces(fig)
+
+            # Finite trades keep their relative endpoints while the unknown trade uses the minima.
+            np.testing.assert_array_equal(nt["Closed - Profit"].marker.size, [7.0, 14.0])
+            np.testing.assert_array_equal(nt["Closed - Profit"].marker.opacity, [0.75, 0.9])
+            np.testing.assert_array_equal(nt["Open"].marker.size, [7.0])
+            np.testing.assert_array_equal(nt["Open"].marker.opacity, [0.75])
+
+            # Guard every emitted trace against metadata that Plotly would reject.
+            for trace in fig.data:
+                assert np.all(np.isfinite(trace.marker.size))
+                assert np.all(np.isfinite(trace.marker.opacity))
+
 
 class TestDrawdownsPlot:
     def test_drawdowns_plot_with_zones(self):

vectorbt/portfolio/trades.py

@@ -910,8 +910,17 @@ def plot_pnl(
             profit_mask = pnl > 0
             loss_mask = pnl < 0
 
-            marker_size = min_rel_rescale(np.abs(returns), marker_size_range)
-            opacity = max_rel_rescale(np.abs(returns), opacity_range)
+            # Exclude NaN and infinity so they cannot contaminate scaling for every trade marker.
+            abs_returns = np.abs(returns)
+            finite_mask = np.isfinite(abs_returns)
+
+            # Start unknown returns at the lower bounds, which are valid Plotly marker values.
+            marker_size = np.full(abs_returns.shape, marker_size_range[0], dtype=float)
+            opacity = np.full(abs_returns.shape, opacity_range[0], dtype=float)
+            if np.any(finite_mask):
+                # Preserve the existing relative scale for finite trades only.
+                marker_size[finite_mask] = min_rel_rescale(abs_returns[finite_mask], marker_size_range)
+                opacity[finite_mask] = max_rel_rescale(abs_returns[finite_mask], opacity_range)
 
             open_mask = status == TradeStatus.Open
             closed_profit_mask = (~open_mask) & profit_mask