Anto01
1612991d0d
Introduces a new plugin architecture for study-specific physics visualizations, separating "optimizer perspective" (Analysis) from "engineer perspective" (Insights). New module: optimization_engine/insights/ - base.py: StudyInsight base class, InsightConfig, InsightResult, registry - zernike_wfe.py: Mirror WFE with 3D surface and Zernike decomposition - stress_field.py: Von Mises stress contours with safety factors - modal_analysis.py: Natural frequencies and mode shapes - thermal_field.py: Temperature distribution visualization - design_space.py: Parameter-objective landscape exploration Features: - 5 insight types: zernike_wfe, stress_field, modal, thermal, design_space - CLI: python -m optimization_engine.insights generate <study> - Standalone HTML generation with Plotly - Enhanced Zernike viz: Turbo colorscale, smooth shading, 0.5x AMP - Dashboard API fix: Added include_coefficients param to extract_relative() Documentation: - docs/protocols/system/SYS_16_STUDY_INSIGHTS.md - Updated ATOMIZER_CONTEXT.md (v1.7) - Updated 01_CHEATSHEET.md with insights section Tools: - tools/zernike_html_generator.py: Standalone WFE HTML generator - tools/analyze_wfe.bat: Double-click to analyze OP2 files 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
373 lines
12 KiB
Python
373 lines
12 KiB
Python
"""
|
|
Design Space Insight
|
|
|
|
Provides interactive visualization of the design space explored during optimization.
|
|
Shows parameter relationships, objective landscapes, and design evolution.
|
|
|
|
This insight bridges optimization metrics (from Analysis) with physics understanding,
|
|
showing how design parameters affect the physical objectives.
|
|
|
|
Applicable to: All optimization studies with completed trials.
|
|
"""
|
|
|
|
from pathlib import Path
|
|
from datetime import datetime
|
|
from typing import Dict, Any, List, Optional, Tuple
|
|
import sqlite3
|
|
import json
|
|
import numpy as np
|
|
|
|
from .base import StudyInsight, InsightConfig, InsightResult, register_insight
|
|
|
|
# Lazy imports
|
|
_plotly_loaded = False
|
|
_go = None
|
|
_make_subplots = None
|
|
|
|
|
|
def _load_dependencies():
|
|
"""Lazy load heavy dependencies."""
|
|
global _plotly_loaded, _go, _make_subplots
|
|
if not _plotly_loaded:
|
|
import plotly.graph_objects as go
|
|
from plotly.subplots import make_subplots
|
|
_go = go
|
|
_make_subplots = make_subplots
|
|
_plotly_loaded = True
|
|
|
|
|
|
@register_insight
|
|
class DesignSpaceInsight(StudyInsight):
|
|
"""
|
|
Design space exploration visualization.
|
|
|
|
Shows:
|
|
- Parallel coordinates plot of parameters vs objectives
|
|
- Scatter matrix of parameter relationships
|
|
- 3D parameter-objective landscape
|
|
- Best design summary with physics interpretation
|
|
"""
|
|
|
|
insight_type = "design_space"
|
|
name = "Design Space Explorer"
|
|
description = "Interactive parameter-objective relationship visualization"
|
|
applicable_to = ["all"] # Works with any optimization study
|
|
required_files = [] # Requires study.db, not OP2
|
|
|
|
def __init__(self, study_path: Path):
|
|
super().__init__(study_path)
|
|
self.db_path = self.results_path / "study.db"
|
|
self._trials: Optional[List[Dict]] = None
|
|
self._params: Optional[List[str]] = None
|
|
self._objectives: Optional[List[str]] = None
|
|
|
|
def can_generate(self) -> bool:
|
|
"""Check if study.db exists with trial data."""
|
|
if not self.db_path.exists():
|
|
return False
|
|
|
|
try:
|
|
conn = sqlite3.connect(str(self.db_path))
|
|
cursor = conn.cursor()
|
|
cursor.execute("SELECT COUNT(*) FROM trials WHERE state = 'COMPLETE'")
|
|
count = cursor.fetchone()[0]
|
|
conn.close()
|
|
return count >= 5 # Need at least 5 trials
|
|
except Exception:
|
|
return False
|
|
|
|
def _load_data(self):
|
|
"""Load trial data from study.db."""
|
|
if self._trials is not None:
|
|
return
|
|
|
|
conn = sqlite3.connect(str(self.db_path))
|
|
cursor = conn.cursor()
|
|
|
|
# Get completed trials
|
|
cursor.execute("""
|
|
SELECT trial_id, params, values, state
|
|
FROM trials
|
|
WHERE state = 'COMPLETE'
|
|
ORDER BY trial_id
|
|
""")
|
|
|
|
self._trials = []
|
|
self._params = None
|
|
self._objectives = None
|
|
|
|
for row in cursor.fetchall():
|
|
trial_id, params_json, values_json, state = row
|
|
|
|
try:
|
|
params = json.loads(params_json) if params_json else {}
|
|
values = json.loads(values_json) if values_json else {}
|
|
except json.JSONDecodeError:
|
|
continue
|
|
|
|
# Flatten nested values
|
|
flat_values = {}
|
|
for k, v in values.items():
|
|
if isinstance(v, dict):
|
|
flat_values.update(v)
|
|
else:
|
|
flat_values[k] = v
|
|
|
|
self._trials.append({
|
|
'trial_id': trial_id,
|
|
'params': params,
|
|
'values': flat_values,
|
|
})
|
|
|
|
# Extract param and objective names from first trial
|
|
if self._params is None:
|
|
self._params = list(params.keys())
|
|
if self._objectives is None:
|
|
self._objectives = list(flat_values.keys())
|
|
|
|
conn.close()
|
|
|
|
def _generate(self, config: InsightConfig) -> InsightResult:
|
|
"""Generate design space visualization."""
|
|
self._load_data()
|
|
|
|
if not self._trials or len(self._trials) < 5:
|
|
return InsightResult(success=False,
|
|
error=f"Need at least 5 trials, found: {len(self._trials or [])}")
|
|
|
|
_load_dependencies()
|
|
|
|
# Configuration
|
|
colorscale = config.extra.get('colorscale', 'Viridis')
|
|
primary_objective = config.extra.get('primary_objective', None)
|
|
|
|
# Use first objective if not specified
|
|
if primary_objective is None and self._objectives:
|
|
primary_objective = self._objectives[0]
|
|
|
|
# Build data arrays
|
|
n_trials = len(self._trials)
|
|
param_data = {p: [] for p in self._params}
|
|
obj_data = {o: [] for o in self._objectives}
|
|
trial_ids = []
|
|
|
|
for trial in self._trials:
|
|
trial_ids.append(trial['trial_id'])
|
|
for p in self._params:
|
|
param_data[p].append(trial['params'].get(p, np.nan))
|
|
for o in self._objectives:
|
|
obj_data[o].append(trial['values'].get(o, np.nan))
|
|
|
|
# Convert to arrays
|
|
for p in self._params:
|
|
param_data[p] = np.array(param_data[p])
|
|
for o in self._objectives:
|
|
obj_data[o] = np.array(obj_data[o])
|
|
|
|
# Find best trial
|
|
if primary_objective and primary_objective in obj_data:
|
|
obj_values = obj_data[primary_objective]
|
|
valid_mask = ~np.isnan(obj_values)
|
|
if np.any(valid_mask):
|
|
best_idx = np.nanargmin(obj_values)
|
|
best_trial = self._trials[best_idx]
|
|
best_value = obj_values[best_idx]
|
|
else:
|
|
best_trial = None
|
|
best_value = None
|
|
else:
|
|
best_trial = None
|
|
best_value = None
|
|
|
|
# Build visualization
|
|
n_params = len(self._params)
|
|
n_objs = len(self._objectives)
|
|
|
|
# Layout: 2x2 grid
|
|
fig = _make_subplots(
|
|
rows=2, cols=2,
|
|
specs=[
|
|
[{"type": "parcoords", "colspan": 2}, None],
|
|
[{"type": "scatter3d" if n_params >= 2 else "xy"},
|
|
{"type": "table"}]
|
|
],
|
|
row_heights=[0.55, 0.45],
|
|
subplot_titles=[
|
|
"<b>Parallel Coordinates - Design Space</b>",
|
|
"<b>Parameter Landscape</b>",
|
|
"<b>Best Design</b>"
|
|
]
|
|
)
|
|
|
|
# 1. Parallel coordinates
|
|
dimensions = []
|
|
|
|
# Add parameters
|
|
for p in self._params:
|
|
values = param_data[p]
|
|
if not np.all(np.isnan(values)):
|
|
dimensions.append(dict(
|
|
label=p,
|
|
values=values,
|
|
range=[float(np.nanmin(values)), float(np.nanmax(values))]
|
|
))
|
|
|
|
# Add objectives
|
|
for o in self._objectives:
|
|
values = obj_data[o]
|
|
if not np.all(np.isnan(values)):
|
|
dimensions.append(dict(
|
|
label=o,
|
|
values=values,
|
|
range=[float(np.nanmin(values)), float(np.nanmax(values))]
|
|
))
|
|
|
|
if dimensions:
|
|
# Color by primary objective
|
|
color_values = obj_data.get(primary_objective, trial_ids)
|
|
if isinstance(color_values, list):
|
|
color_values = np.array(color_values)
|
|
|
|
fig.add_trace(_go.Parcoords(
|
|
line=dict(
|
|
color=color_values,
|
|
colorscale=colorscale,
|
|
showscale=True,
|
|
colorbar=dict(title=primary_objective or "Trial", thickness=15)
|
|
),
|
|
dimensions=dimensions,
|
|
), row=1, col=1)
|
|
|
|
# 2. 3D Parameter landscape (first 2 params vs primary objective)
|
|
if n_params >= 2 and primary_objective:
|
|
x_param = self._params[0]
|
|
y_param = self._params[1]
|
|
z_values = obj_data.get(primary_objective, [])
|
|
|
|
fig.add_trace(_go.Scatter3d(
|
|
x=param_data[x_param],
|
|
y=param_data[y_param],
|
|
z=z_values,
|
|
mode='markers',
|
|
marker=dict(
|
|
size=6,
|
|
color=z_values,
|
|
colorscale=colorscale,
|
|
opacity=0.8,
|
|
showscale=False,
|
|
),
|
|
text=[f"Trial {tid}" for tid in trial_ids],
|
|
hovertemplate=(
|
|
f"{x_param}: %{{x:.3f}}<br>"
|
|
f"{y_param}: %{{y:.3f}}<br>"
|
|
f"{primary_objective}: %{{z:.4f}}<br>"
|
|
"%{text}<extra></extra>"
|
|
),
|
|
), row=2, col=1)
|
|
|
|
# Highlight best point
|
|
if best_trial:
|
|
fig.add_trace(_go.Scatter3d(
|
|
x=[best_trial['params'].get(x_param)],
|
|
y=[best_trial['params'].get(y_param)],
|
|
z=[best_value],
|
|
mode='markers',
|
|
marker=dict(size=12, color='red', symbol='diamond'),
|
|
name='Best',
|
|
showlegend=True,
|
|
), row=2, col=1)
|
|
|
|
fig.update_scenes(
|
|
xaxis_title=x_param,
|
|
yaxis_title=y_param,
|
|
zaxis_title=primary_objective,
|
|
)
|
|
elif n_params >= 1 and primary_objective:
|
|
# 2D scatter
|
|
x_param = self._params[0]
|
|
z_values = obj_data.get(primary_objective, [])
|
|
|
|
fig.add_trace(_go.Scatter(
|
|
x=param_data[x_param],
|
|
y=z_values,
|
|
mode='markers',
|
|
marker=dict(size=8, color=z_values, colorscale=colorscale),
|
|
), row=2, col=1)
|
|
|
|
fig.update_xaxes(title_text=x_param, row=2, col=1)
|
|
fig.update_yaxes(title_text=primary_objective, row=2, col=1)
|
|
|
|
# 3. Best design table
|
|
if best_trial:
|
|
labels = ["<b>Metric</b>", "<b>Value</b>"]
|
|
|
|
# Combine params and objectives
|
|
table_labels = ["Trial ID"] + self._params + self._objectives
|
|
table_values = [str(best_trial['trial_id'])]
|
|
|
|
for p in self._params:
|
|
val = best_trial['params'].get(p, 'N/A')
|
|
table_values.append(f"{val:.4f}" if isinstance(val, (int, float)) else str(val))
|
|
|
|
for o in self._objectives:
|
|
val = best_trial['values'].get(o, 'N/A')
|
|
table_values.append(f"{val:.4f}" if isinstance(val, (int, float)) else str(val))
|
|
|
|
fig.add_trace(_go.Table(
|
|
header=dict(values=labels,
|
|
fill_color='#1f2937', font=dict(color='white')),
|
|
cells=dict(values=[table_labels, table_values],
|
|
fill_color='#374151', font=dict(color='white'))
|
|
), row=2, col=2)
|
|
else:
|
|
fig.add_trace(_go.Table(
|
|
header=dict(values=["<b>Info</b>"],
|
|
fill_color='#1f2937', font=dict(color='white')),
|
|
cells=dict(values=[["No valid trials found"]],
|
|
fill_color='#374151', font=dict(color='white'))
|
|
), row=2, col=2)
|
|
|
|
# Layout
|
|
fig.update_layout(
|
|
width=1500, height=1000,
|
|
paper_bgcolor='#111827', plot_bgcolor='#1f2937',
|
|
font=dict(color='white'),
|
|
title=dict(
|
|
text=f"<b>Atomizer Design Space Explorer</b><br>"
|
|
f"<sub>{n_trials} trials, {n_params} parameters, {n_objs} objectives</sub>",
|
|
x=0.5, font=dict(size=18)
|
|
),
|
|
)
|
|
|
|
# Save HTML
|
|
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
|
|
output_dir = config.output_dir or self.insights_path
|
|
output_dir.mkdir(parents=True, exist_ok=True)
|
|
|
|
html_path = output_dir / f"design_space_{timestamp}.html"
|
|
html_path.write_text(
|
|
fig.to_html(include_plotlyjs='cdn', full_html=True),
|
|
encoding='utf-8'
|
|
)
|
|
|
|
# Summary
|
|
summary = {
|
|
'n_trials': n_trials,
|
|
'n_params': n_params,
|
|
'n_objectives': n_objs,
|
|
'parameters': self._params,
|
|
'objectives': self._objectives,
|
|
}
|
|
|
|
if best_trial:
|
|
summary['best_trial_id'] = best_trial['trial_id']
|
|
summary['best_params'] = best_trial['params']
|
|
summary['best_values'] = best_trial['values']
|
|
|
|
return InsightResult(
|
|
success=True,
|
|
html_path=html_path,
|
|
plotly_figure=fig.to_dict(),
|
|
summary=summary
|
|
)
|