"""Visualization for value-binned error analysis."""
import numpy as np
import matplotlib.pyplot as plt
from matplotlib.gridspec import GridSpecFromSubplotSpec
[文档]
def plot_rearranged_error_by_value(results: dict, axes=None, ax=None, title: str = None):
"""
Plot value-binned mean residual and RMS residual.
Creates a comprehensive visualization showing:
- Top panel: Scatter of raw residual vs value bin with mean residual overlay
- Bottom panel: RMS residual vs value bin as bar chart
Parameters
----------
results : dict
Dictionary from rearrange_error_by_value().
axes : tuple or array, optional
Tuple of (ax1, ax2) for top and bottom panels.
ax : matplotlib.axes.Axes, optional
Single axis to split into 2 panels.
title : str, optional
Test setup description for title.
"""
# --- 1. Extract Data (Using new simplified keys) ---
error = results.get('error', np.array([]))
bin_indices = results.get('bin_indices', np.array([]))
error_mean = results['error_mean']
error_rms = results['error_rms']
bin_centers = results['bin_centers']
value_bin_centers = results.get('value_bin_centers', bin_centers)
n_bins = results['n_bins']
scatter_x = value_bin_centers[bin_indices]
finite_value_bins = np.isfinite(value_bin_centers)
if len(bin_indices) == 0: return
# --- 2. Axes Management ---
if axes is not None:
# Support both tuple (ax1, ax2) and numpy array [ax1, ax2]
# This makes it compatible with axes = plt.subplots(2, 1)[1]
ax1, ax2 = axes if isinstance(axes, (tuple, list)) else axes.flatten()
else:
# Single axis (or None), get current axis and split it
if ax is None:
ax = plt.gca()
# Split single axis into 2 rows
fig = ax.get_figure()
if hasattr(ax, 'get_subplotspec') and ax.get_subplotspec():
gs = GridSpecFromSubplotSpec(2, 1, subplot_spec=ax.get_subplotspec(), hspace=0.35)
ax.remove() # Remove original placeholder axis
ax1 = fig.add_subplot(gs[0])
ax2 = fig.add_subplot(gs[1])
else:
# Fallback for manual positioning
pos = ax.get_position()
ax.remove()
ax1 = fig.add_axes([pos.x0, pos.y0 + pos.height/2, pos.width, pos.height/2])
ax2 = fig.add_axes([pos.x0, pos.y0, pos.width, pos.height/2])
# ======================================================================
# Top Panel: Scatter of error vs bin index with mean error overlay
# ======================================================================
if len(error) > 0:
# Scatter plot: High transparency, rasterized for performance on large datasets
ax1.scatter(scatter_x, error, alpha=0.2, s=1, color='red',
rasterized=True, label='Raw Residual')
# Mean residual line overlay (value-binned conditional mean)
ax1.plot(value_bin_centers, error_mean, 'b-', linewidth=1.5,
label='Value-Binned Mean Residual')
# Set axis limits
if np.any(finite_value_bins):
x_min = np.nanmin(value_bin_centers)
x_max = np.nanmax(value_bin_centers)
x_margin = 0.02 * (x_max - x_min) if x_max > x_min else 0.5
ax1.set_xlim([x_min - x_margin, x_max + x_margin])
# Smart Y-limits
y_min, y_max = np.min(error), np.max(error)
y_range = y_max - y_min
margin = y_range * 0.1 if y_range != 0 else 1.0
ax1.set_ylim([y_min - margin, y_max + margin])
ax1.set_ylabel('Residual')
if title:
ax1.set_title(title)
else:
ax1.set_title('Value-Binned Residual Diagnostic')
ax1.grid(True, alpha=0.3)
ax1.legend(loc='upper right', fontsize=8)
ax1.set_xlabel('Signal Value')
# ======================================================================
# Bottom Panel: RMS Error Bar Chart (Noise Profile)
# ======================================================================
if len(bin_centers) > 0:
# Bar chart
if len(value_bin_centers) > 1:
bar_width = 0.9 * np.nanmedian(np.diff(value_bin_centers))
else:
bar_width = 0.9
ax2.bar(value_bin_centers, error_rms, width=bar_width,
color='skyblue', alpha=0.8, edgecolor='darkblue', linewidth=0.3)
# Set axis limits
if np.any(finite_value_bins):
x_min = np.nanmin(value_bin_centers)
x_max = np.nanmax(value_bin_centers)
x_margin = 0.02 * (x_max - x_min) if x_max > x_min else 0.5
ax2.set_xlim([x_min - x_margin, x_max + x_margin])
ax2.set_ylim([0, np.nanmax(error_rms) * 1.15])
ax2.set_xlabel('Signal Value')
ax2.set_ylabel('RMS Residual')
ax2.set_title('Value-Binned RMS Residual')
ax2.grid(True, alpha=0.3, axis='y')
# Tight layout if we created the figure structure ourselves
if ax is None:
plt.tight_layout()
