Spaces:

2catycm
/

VisualizationForGeneralizedGaussianMixture

Sleeping

VisualizationForGeneralizedGaussianMixture / experiments /performance_comparison.py

feat: 基本画完，缺乏大小标尺

4a7325a over 1 year ago

5.27 kB

import plotly.graph_objects as go
import pandas as pd
import numpy as np

# 生成符合学术规范的模拟数据
np.random.seed(42)  # 确保可重复性
methods = ['RSM', 'PSRN', 'NGGP', 'PySR', 'BMS', 'uDSR', 'AIF', 
          'DGSR', 'E2E', 'SymINDy', 'physo', 'TPSR', 'SPL', 
          'DEAP', 'SINDy', 'NSRS', 'gplearn', 'SNIP', 'KAN', 'EQL']

# 生成0-1之间的RMSE数据（保持原图分布模式）
rmse_values = np.clip(np.abs(np.random.normal(0.3, 0.15, len(methods))), 0.05, 0.95)
uncertainty = np.random.uniform(0.02, 0.08, len(methods))
param_sizes = np.array([1e3, 1e4, 5e4, 1e5, 5e5, 1e6, 2e6])  # 定义标准参数规模

# 构建分类系统（基于方法原理）
method_categories = {
    'Symbolic': ['RSM', 'PySR', 'SymINDy', 'gplearn', 'EQL'],
    'Neural': ['NGGP', 'DGSR', 'E2E', 'KAN'],
    'Evolutionary': ['PSRN', 'DEAP', 'NSRS'],
    'Physics-based': ['physo', 'TPSR', 'SPL'],
    'Hybrid': ['BMS', 'uDSR', 'AIF', 'SINDy', 'SNIP']
}

# 创建数据框架
df = pd.DataFrame({
    'Method': methods,
    'RMSE': rmse_values,
    'Uncertainty': uncertainty,
    'ParamSize': np.random.choice(param_sizes, len(methods))
}).sort_values('RMSE', ascending=True)

# 添加分类信息
df['Category'] = df['Method'].apply(lambda x: next((k for k,v in method_categories.items() if x in v), 'Other'))

# 颜色映射系统（学术级调色板）
category_colors = {
    'Symbolic': '#1f77b4',
    'Neural': '#ff7f0e',
    'Evolutionary': '#2ca02c',
    'Physics-based': '#d62728',
    'Hybrid': '#9467bd'
}

# 创建基础图表
fig = go.Figure()

# 动态尺寸计算系统
size_min = np.log(df['ParamSize'].min())
size_max = np.log(df['ParamSize'].max())
sizes = 15 + 25 * (np.log(df['ParamSize']) - size_min) / (size_max - size_min)  # 动态尺寸范围

# 添加主数据轨迹
for category in df['Category'].unique():
    df_sub = df[df['Category'] == category]
    fig.add_trace(go.Scatter(
        x=df_sub['RMSE'],
        y=df_sub['Method'],
        mode='markers',
        name=category,
        marker=dict(
            size=sizes[df_sub.index],
            color=category_colors[category],
            opacity=0.9,
            line=dict(width=1, color='black')
        ),
        error_x=dict(
            type='data',
            array=df_sub['Uncertainty'],
            color='rgba(40,40,40,0.6)',
            thickness=1.2,
            width=10
        ),
        hoverinfo='text',
        hovertext=df_sub.apply(lambda r: f"{r['Method']}<br>RMSE: {r['RMSE']:.3f} ± {r['Uncertainty']:.3f}<br>Params: {r['ParamSize']:,.0f}", axis=1)
    ))

# 动态轴范围计算
data_min = (df['RMSE'] - df['Uncertainty']).min()
x_min = max(data_min - 0.05, 0)  # 保证最小值不低于0
x_max = min(df['RMSE'].max() + df['Uncertainty'].max(), 1)  # 保证最大值不超过1

# 专业级布局配置
fig.update_layout(
    title='Methods RMSE Comparison with Parameter Scale',
    xaxis=dict(
        title='Root Mean Square Error (RMSE) → Lower is better',
        range=[x_min, x_max],
        tickvals=np.arange(0, 1.1, 0.1),
        gridcolor='#F0F0F0',
        zeroline=False,
        showspikes=True
    ),
    yaxis=dict(
        categoryorder='array',
        categoryarray=df['Method'].tolist(),
        tickfont=dict(size=12),
        showticklabels=False  # 禁用默认标签
    ),
    plot_bgcolor='white',
    width=1100,
    height=700,
    margin=dict(l=180, r=50, t=80, b=40),
    legend=dict(
        title='Method Categories',
        orientation='v',
        yanchor="top",
        y=0.98,
        xanchor="left",
        x=1.02
    )
)

# 添加自定义y轴标签（分类着色）
y_positions = np.linspace(0.03, 0.97, len(methods))  # 动态计算标签位置
for idx, method in enumerate(df['Method']):
    category = df[df['Method'] == method]['Category'].values[0]
    fig.add_annotation(
        x=0.01,
        y=y_positions[idx],
        xref='paper',
        yref='paper',
        text=method,
        showarrow=False,
        font=dict(
            size=12,
            color=category_colors[category]
        ),
        xanchor='right'
    )

# 添加专业级尺寸图例系统
# size_legend_values = [1e3, 1e4, 1e5, 1e6]  # 定义标准参数规模
# size_legend_sizes = 15 + 25 * (np.log(size_legend_values) - size_min) / (size_max - size_min)

# fig.add_trace(go.Scatter(
#     x=[0.52, 0.55, 0.58, 0.61],
#     y=np.array([0.00, 0.05, 0.10, 0.15]),
#     mode='markers',
#     marker=dict(
#         size=size_legend_sizes,
#         color='#444444',
#         opacity=0.8
#     ),
#     showlegend=False,
#     text=[f'{size:.2e}' for size in size_legend_values],
# ))

# # 添加尺寸图例标注
# size_labels = ['1K', '10K', '100K', '1M']
# for i, (x, y, label) in enumerate(zip([0.95]*4, [0.15,0.20,0.25,0.30], size_labels)):
#     fig.add_annotation(
#         x=x,
#         y=y,
#         xref="paper",
#         yref="paper",
#         text=label,
#         showarrow=False,
#         font=dict(size=10),
#         xanchor='left'
#     )

# | Parameters (log scale)

# 添加最终标注
fig.add_annotation(
    x=0.98, y=0.02,
    xref='paper', yref='paper',
    text='叶璨铭绘制',
    showarrow=False,
    font=dict(size=10, color='#666666'),
    bgcolor='white'
)

fig.show()

	import plotly.graph_objects as go
	import pandas as pd
	import numpy as np

	# 生成符合学术规范的模拟数据
	np.random.seed(42) # 确保可重复性
	methods = ['RSM', 'PSRN', 'NGGP', 'PySR', 'BMS', 'uDSR', 'AIF',
	'DGSR', 'E2E', 'SymINDy', 'physo', 'TPSR', 'SPL',
	'DEAP', 'SINDy', 'NSRS', 'gplearn', 'SNIP', 'KAN', 'EQL']

	# 生成0-1之间的RMSE数据（保持原图分布模式）
	rmse_values = np.clip(np.abs(np.random.normal(0.3, 0.15, len(methods))), 0.05, 0.95)
	uncertainty = np.random.uniform(0.02, 0.08, len(methods))
	param_sizes = np.array([1e3, 1e4, 5e4, 1e5, 5e5, 1e6, 2e6]) # 定义标准参数规模

	# 构建分类系统（基于方法原理）
	method_categories = {
	'Symbolic': ['RSM', 'PySR', 'SymINDy', 'gplearn', 'EQL'],
	'Neural': ['NGGP', 'DGSR', 'E2E', 'KAN'],
	'Evolutionary': ['PSRN', 'DEAP', 'NSRS'],
	'Physics-based': ['physo', 'TPSR', 'SPL'],
	'Hybrid': ['BMS', 'uDSR', 'AIF', 'SINDy', 'SNIP']
	}

	# 创建数据框架
	df = pd.DataFrame({
	'Method': methods,
	'RMSE': rmse_values,
	'Uncertainty': uncertainty,
	'ParamSize': np.random.choice(param_sizes, len(methods))
	}).sort_values('RMSE', ascending=True)

	# 添加分类信息
	df['Category'] = df['Method'].apply(lambda x: next((k for k,v in method_categories.items() if x in v), 'Other'))

	# 颜色映射系统（学术级调色板）
	category_colors = {
	'Symbolic': '#1f77b4',
	'Neural': '#ff7f0e',
	'Evolutionary': '#2ca02c',
	'Physics-based': '#d62728',
	'Hybrid': '#9467bd'
	}

	# 创建基础图表
	fig = go.Figure()

	# 动态尺寸计算系统
	size_min = np.log(df['ParamSize'].min())
	size_max = np.log(df['ParamSize'].max())
	sizes = 15 + 25 * (np.log(df['ParamSize']) - size_min) / (size_max - size_min) # 动态尺寸范围

	# 添加主数据轨迹
	for category in df['Category'].unique():
	df_sub = df[df['Category'] == category]
	fig.add_trace(go.Scatter(
	x=df_sub['RMSE'],
	y=df_sub['Method'],
	mode='markers',
	name=category,
	marker=dict(
	size=sizes[df_sub.index],
	color=category_colors[category],
	opacity=0.9,
	line=dict(width=1, color='black')
	),
	error_x=dict(
	type='data',
	array=df_sub['Uncertainty'],
	color='rgba(40,40,40,0.6)',
	thickness=1.2,
	width=10
	),
	hoverinfo='text',
	hovertext=df_sub.apply(lambda r: f"{r['Method']}<br>RMSE: {r['RMSE']:.3f} ± {r['Uncertainty']:.3f}<br>Params: {r['ParamSize']:,.0f}", axis=1)
	))

	# 动态轴范围计算
	data_min = (df['RMSE'] - df['Uncertainty']).min()
	x_min = max(data_min - 0.05, 0) # 保证最小值不低于0
	x_max = min(df['RMSE'].max() + df['Uncertainty'].max(), 1) # 保证最大值不超过1

	# 专业级布局配置
	fig.update_layout(
	title='Methods RMSE Comparison with Parameter Scale',
	xaxis=dict(
	title='Root Mean Square Error (RMSE) → Lower is better',
	range=[x_min, x_max],
	tickvals=np.arange(0, 1.1, 0.1),
	gridcolor='#F0F0F0',
	zeroline=False,
	showspikes=True
	),
	yaxis=dict(
	categoryorder='array',
	categoryarray=df['Method'].tolist(),
	tickfont=dict(size=12),
	showticklabels=False # 禁用默认标签
	),
	plot_bgcolor='white',
	width=1100,
	height=700,
	margin=dict(l=180, r=50, t=80, b=40),
	legend=dict(
	title='Method Categories',
	orientation='v',
	yanchor="top",
	y=0.98,
	xanchor="left",
	x=1.02
	)
	)

	# 添加自定义y轴标签（分类着色）
	y_positions = np.linspace(0.03, 0.97, len(methods)) # 动态计算标签位置
	for idx, method in enumerate(df['Method']):
	category = df[df['Method'] == method]['Category'].values[0]
	fig.add_annotation(
	x=0.01,
	y=y_positions[idx],
	xref='paper',
	yref='paper',
	text=method,
	showarrow=False,
	font=dict(
	size=12,
	color=category_colors[category]
	),
	xanchor='right'
	)

	# 添加专业级尺寸图例系统
	# size_legend_values = [1e3, 1e4, 1e5, 1e6] # 定义标准参数规模
	# size_legend_sizes = 15 + 25 * (np.log(size_legend_values) - size_min) / (size_max - size_min)

	# fig.add_trace(go.Scatter(
	# x=[0.52, 0.55, 0.58, 0.61],
	# y=np.array([0.00, 0.05, 0.10, 0.15]),
	# mode='markers',
	# marker=dict(
	# size=size_legend_sizes,
	# color='#444444',
	# opacity=0.8
	# ),
	# showlegend=False,
	# text=[f'{size:.2e}' for size in size_legend_values],
	# ))

	# # 添加尺寸图例标注
	# size_labels = ['1K', '10K', '100K', '1M']
	# for i, (x, y, label) in enumerate(zip([0.95]*4, [0.15,0.20,0.25,0.30], size_labels)):
	# fig.add_annotation(
	# x=x,
	# y=y,
	# xref="paper",
	# yref="paper",
	# text=label,
	# showarrow=False,
	# font=dict(size=10),
	# xanchor='left'
	# )

	# \| Parameters (log scale)

	# 添加最终标注
	fig.add_annotation(
	x=0.98, y=0.02,
	xref='paper', yref='paper',
	text='叶璨铭绘制',
	showarrow=False,
	font=dict(size=10, color='#666666'),
	bgcolor='white'
	)

	fig.show()