import plotly.graph_objects as go
import plotly.express as px
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib
matplotlib.use('Agg') # 使用非互動式後端
import arviz as az
import io
import base64
from PIL import Image
def plot_trace(trace, var_names=['d', 'sigma']):
"""
繪製 Trace Plot（MCMC 收斂診斷）
包含完整的 warmup + posterior
Args:
trace: ArviZ InferenceData 物件
var_names: 要繪製的變數名稱
Returns:
PIL Image
"""
fig, axes = plt.subplots(len(var_names), 2, figsize=(14, 4 * len(var_names)))
if len(var_names) == 1:
axes = axes.reshape(1, -1)
# 檢查是否有 warmup_posterior
has_warmup = hasattr(trace, 'warmup_posterior') and trace.warmup_posterior is not None
for idx, var_name in enumerate(var_names):
# 左圖: KDE 密度圖（只用 posterior, 不用 warmup）
post_data = trace.posterior[var_name].values
for chain_idx in range(post_data.shape[0]):
from scipy import stats
data = post_data[chain_idx].flatten()
density = stats.gaussian_kde(data)
xs = np.linspace(data.min(), data.max(), 200)
axes[idx, 0].plot(xs, density(xs), alpha=0.8, label=f'Chain {chain_idx+1}')
axes[idx, 0].set_xlabel(var_name, fontsize=12)
axes[idx, 0].set_ylabel('Density', fontsize=12)
axes[idx, 0].set_title(f'{var_name}', fontsize=13, fontweight='bold')
if idx == 0:
axes[idx, 0].legend()
# 右圖: Trace 圖（完整 warmup + posterior）
if has_warmup:
# 有 warmup: 合併繪製
warmup_data = trace.warmup_posterior[var_name].values
post_data = trace.posterior[var_name].values
n_warmup = warmup_data.shape[1]
n_post = post_data.shape[1]
# 定義顏色,讓每條鏈用固定顏色
colors = plt.cm.tab10.colors # 使用 matplotlib 的顏色循環
for chain_idx in range(warmup_data.shape[0]):
chain_color = colors[chain_idx % len(colors)] # 每條鏈一個固定顏色
# 繪 warmup 部分
x_warmup = np.arange(n_warmup)
axes[idx, 1].plot(x_warmup, warmup_data[chain_idx].flatten(),
color=chain_color, # 👈 指定顏色
alpha=0.7, linewidth=0.5,
label=f'Chain {chain_idx+1}' if idx == 0 else '')
# 繪 posterior 部分 (用同樣的顏色!)
x_post = np.arange(n_warmup, n_warmup + n_post)
axes[idx, 1].plot(x_post, post_data[chain_idx].flatten(),
color=chain_color, # 👈 同一個顏色
alpha=0.7, linewidth=0.5)
# 加 Tune 結束的紅線
axes[idx, 1].axvline(x=n_warmup, color='red', linestyle='--',
linewidth=2, alpha=0.7,
label='Tune結束' if idx == 0 else '')
else:
# 沒有 warmup: 只用 posterior
post_data = trace.posterior[var_name].values
for chain_idx in range(post_data.shape[0]):
axes[idx, 1].plot(post_data[chain_idx].flatten(),
alpha=0.7, linewidth=0.5,
label=f'Chain {chain_idx+1}' if idx == 0 else '')
axes[idx, 1].set_xlabel('Iteration', fontsize=12)
axes[idx, 1].set_ylabel(var_name, fontsize=12)
axes[idx, 1].set_title(f'{var_name} trace', fontsize=13, fontweight='bold')
if idx == 0:
axes[idx, 1].legend(loc='upper right', fontsize=9)
axes[idx, 1].grid(alpha=0.3)
plt.tight_layout()
# 轉換為圖片
buf = io.BytesIO()
plt.savefig(buf, format='png', dpi=300, bbox_inches='tight')
buf.seek(0)
img = Image.open(buf)
plt.close()
return img
# ============================================
# 替換說明:
# 在 bayesian_utils.py 中,把第 13-51 行的整個 plot_trace 函數
# 替換成上面這個版本
# ============================================
def plot_posterior(trace, var_names=['d', 'sigma', 'or_speed'], hdi_prob=0.95):
"""
繪製後驗分佈圖
Args:
trace: ArviZ InferenceData 物件
var_names: 要繪製的變數名稱
hdi_prob: HDI 機率
Returns:
PIL Image
"""
fig = az.plot_posterior(trace, var_names=var_names, hdi_prob=hdi_prob, figsize=(14, 5))
plt.tight_layout()
# 轉換為圖片
buf = io.BytesIO()
plt.savefig(buf, format='png', dpi=300, bbox_inches='tight')
buf.seek(0)
img = Image.open(buf)
plt.close()
return img
def plot_forest(trace, trial_labels, title='Effect of Speed on Win Rate by Type'):
"""
繪製 Forest Plot（各屬性效應）
Args:
trace: ArviZ InferenceData 物件
trial_labels: 屬性標籤列表
title: 圖表標題
Returns:
PIL Image
"""
num_trials = len(trial_labels)
# 計算統計量
delta_posterior = trace.posterior['delta'].values.reshape(-1, num_trials)
delta_mean = delta_posterior.mean(axis=0)
delta_hdi = az.hdi(trace, var_names=['delta'], hdi_prob=0.95)['delta'].values
# 建立圖表
fig, ax = plt.subplots(figsize=(12, max(10, num_trials * 0.4)))
y_pos = np.arange(num_trials)
# 繪製信賴區間（橫線）
ax.hlines(y_pos, delta_hdi[:, 0], delta_hdi[:, 1], color='steelblue', linewidth=3, label='95% HDI')
# 繪製平均值（點）
ax.scatter(delta_mean, y_pos, color='darkblue', s=120, zorder=3,
edgecolors='white', linewidth=1.5, label='Mean')
# 標註顯著的點
for i, (mean, hdi) in enumerate(zip(delta_mean, delta_hdi)):
if hdi[0] > 0: # 顯著正效應
ax.text(mean, i, ' ★', fontsize=15, ha='left', va='center', color='gold')
elif hdi[1] < 0: # 顯著負效應
ax.text(mean, i, ' ☆', fontsize=15, ha='left', va='center', color='red')
# 設定軸
ax.set_yticks(y_pos)
ax.set_yticklabels(trial_labels, fontsize=11)
ax.invert_yaxis()
ax.axvline(0, color='red', linestyle='--', linewidth=2, label='No Effect (δ=0)')
ax.set_xlabel('Delta (Log Odds Ratio)', fontsize=13)
ax.set_title(title, fontsize=15, fontweight='bold', pad=20)
ax.legend(loc='lower right')
ax.grid(axis='x', alpha=0.3)
plt.tight_layout()
# 轉換為圖片
buf = io.BytesIO()
plt.savefig(buf, format='png', dpi=300, bbox_inches='tight')
buf.seek(0)
img = Image.open(buf)
plt.close()
return img
def plot_model_dag(analyzer):
"""
繪製模型 DAG 圖
Args:
analyzer: BayesianHierarchicalAnalyzer 物件
Returns:
PIL Image 或 None
"""
try:
gv = analyzer.get_model_graph()
# 轉換為 PNG
png_bytes = gv.pipe(format='png')
# 轉換為 PIL Image
img = Image.open(io.BytesIO(png_bytes))
return img
except Exception as e:
print(f"無法生成 DAG 圖: {e}")
return None
def create_summary_table(results):
"""
創建結果摘要表格
Args:
results: 分析結果字典
Returns:
pandas DataFrame
"""
overall = results['overall']
summary_data = {
'參數': ['d (整體效應)', 'sigma (屬性間變異)', 'or_speed (勝算比)'],
'平均值': [
f"{overall['d_mean']:.4f}",
f"{overall['sigma_mean']:.4f}",
f"{overall['or_mean']:.4f}"
],
'標準差': [
f"{overall['d_sd']:.4f}",
f"{overall['sigma_sd']:.4f}",
f"{overall['or_sd']:.4f}"
],
'95% HDI 下界': [
f"{overall['d_hdi_low']:.4f}",
f"{overall['sigma_hdi_low']:.4f}",
f"{overall['or_hdi_low']:.4f}"
],
'95% HDI 上界': [
f"{overall['d_hdi_high']:.4f}",
f"{overall['sigma_hdi_high']:.4f}",
f"{overall['or_hdi_high']:.4f}"
]
}
return pd.DataFrame(summary_data)
def create_trial_results_table(results):
"""
創建各屬性結果表格
Args:
results: 分析結果字典
Returns:
pandas DataFrame
"""
trial_labels = results['trial_labels']
by_trial = results['by_trial']
data = results['data']
trial_data = {
'屬性': trial_labels,
'Delta (平均)': [f"{x:.4f}" for x in by_trial['delta_mean']],
'Delta (標準差)': [f"{x:.4f}" for x in by_trial['delta_std']],
'95% HDI 下界': [f"{x:.4f}" for x in by_trial['delta_hdi_low']],
'95% HDI 上界': [f"{x:.4f}" for x in by_trial['delta_hdi_high']],
'顯著性': ['★ 顯著' if sig else '不顯著' for sig in by_trial['delta_significant']],
'控制組勝率': [f"{x:.2%}" for x in by_trial['pc_mean']],
'實驗組勝率': [f"{x:.2%}" for x in by_trial['pt_mean']],
'控制組 (勝/總)': [f"{d['rc']}/{d['nc']}" for d in data],
'實驗組 (勝/總)': [f"{d['rt']}/{d['nt']}" for d in data]
}
return pd.DataFrame(trial_data)
def export_results_to_text(results):
"""
匯出結果為純文字格式
Args:
results: 分析結果字典
Returns:
str: 格式化的文字報告
"""
overall = results['overall']
interp = results['interpretation']
diag = results['diagnostics']
report = f"""
==============================================
貝氏階層模型分析報告
==============================================
分析時間: {results['timestamp']}
屬性數量: {results['n_trials']}
----------------------------------------------
1. 整體效應摘要
----------------------------------------------
d (整體效應 - Log OR):
- 平均值: {overall['d_mean']:.4f}
- 標準差: {overall['d_sd']:.4f}
- 95% HDI: [{overall['d_hdi_low']:.4f}, {overall['d_hdi_high']:.4f}]
sigma (屬性間變異):
- 平均值: {overall['sigma_mean']:.4f}
- 標準差: {overall['sigma_sd']:.4f}
- 95% HDI: [{overall['sigma_hdi_low']:.4f}, {overall['sigma_hdi_high']:.4f}]
or_speed (勝算比):
- 平均值: {overall['or_mean']:.4f}
- 標準差: {overall['or_sd']:.4f}
- 95% HDI: [{overall['or_hdi_low']:.4f}, {overall['or_hdi_high']:.4f}]
----------------------------------------------
2. 模型收斂診斷
----------------------------------------------
R-hat (d): {f"{diag['rhat_d']:.4f}" if diag['rhat_d'] is not None else 'N/A'}
R-hat (sigma): {f"{diag['rhat_sigma']:.4f}" if diag['rhat_sigma'] is not None else 'N/A'}
ESS (d): {int(diag['ess_d']) if diag['ess_d'] is not None else 'N/A'}
ESS (sigma): {int(diag['ess_sigma']) if diag['ess_sigma'] is not None else 'N/A'}
收斂狀態: {'✓ 已收斂' if diag['converged'] else '✗ 未收斂'}
----------------------------------------------
3. 結果解釋
----------------------------------------------
整體效應: {interp['overall_effect']}
顯著性: {interp['overall_significance']}
效果大小: {interp['effect_size']}
異質性: {interp['heterogeneity']}
----------------------------------------------
4. 各屬性詳細結果
----------------------------------------------
"""
# 添加各屬性的詳細資訊
trial_labels = results['trial_labels']
by_trial = results['by_trial']
for i, label in enumerate(trial_labels):
sig_marker = "★" if by_trial['delta_significant'][i] else " "
report += f"""
{sig_marker} {label}:
Delta (平均): {by_trial['delta_mean'][i]:.4f}
95% HDI: [{by_trial['delta_hdi_low'][i]:.4f}, {by_trial['delta_hdi_high'][i]:.4f}]
控制組勝率: {by_trial['pc_mean'][i]:.2%}
實驗組勝率: {by_trial['pt_mean'][i]:.2%}
勝率差異: {(by_trial['pt_mean'][i] - by_trial['pc_mean'][i]):.2%}
"""
report += """
==============================================
"""
return report
def plot_odds_ratio_comparison(results):
"""
繪製各屬性的勝算比比較圖（Plotly 版本）
Args:
results: 分析結果字典
Returns:
plotly figure
"""
trial_labels = results['trial_labels']
delta_mean = results['by_trial']['delta_mean']
# 轉換為勝算比
or_values = [np.exp(d) for d in delta_mean]
# 排序
sorted_indices = np.argsort(or_values)[::-1]
sorted_labels = [trial_labels[i] for i in sorted_indices]
sorted_or = [or_values[i] for i in sorted_indices]
sorted_sig = [results['by_trial']['delta_significant'][i] for i in sorted_indices]
# 顏色標記
colors = ['#2ecc71' if sig else '#95a5a6' for sig in sorted_sig]
fig = go.Figure()
fig.add_trace(go.Bar(
x=sorted_or,
y=sorted_labels,
orientation='h',
marker=dict(
color=colors,
line=dict(color='white', width=1)
),
text=[f'{or_val:.2f}' for or_val in sorted_or],
textposition='outside',
hovertemplate='%{y}
OR: %{x:.3f}'
))
# 參考線 (OR = 1)
fig.add_vline(x=1, line_dash="dash", line_color="red", line_width=2)
fig.update_layout(
title='各屬性速度效應（勝算比）',
xaxis_title='Odds Ratio',
yaxis_title='',
width=800,
height=max(400, len(trial_labels) * 25),
template='plotly_white',
showlegend=False
)
return fig