feat: Add Protocol 13 adaptive optimization, Plotly charts, and dashboard improvements
## Protocol 13: Adaptive Multi-Objective Optimization - Iterative FEA + Neural Network surrogate workflow - Initial FEA sampling, NN training, NN-accelerated search - FEA validation of top NN predictions, retraining loop - adaptive_state.json tracks iteration history and best values - M1 mirror study (V11) with 103 FEA, 3000 NN trials ## Dashboard Visualization Enhancements - Added Plotly.js interactive charts (parallel coords, Pareto, convergence) - Lazy loading with React.lazy() for performance - Code splitting: plotly.js-basic-dist (~1MB vs 3.5MB) - Chart library toggle (Recharts default, Plotly on-demand) - ExpandableChart component for full-screen modal views - ConsoleOutput component for real-time log viewing ## Documentation - Protocol 13 detailed documentation - Dashboard visualization guide - Plotly components README - Updated run-optimization skill with Mode 5 (adaptive) ## Bug Fixes - Fixed TypeScript errors in dashboard components - Fixed Card component to accept ReactNode title - Removed unused imports across components 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
Antoine
@@ -83,15 +83,27 @@ class DesignConditionedConv(MessagePassing):
|
||||
Args:
|
||||
x: Node features [num_nodes, in_channels]
|
||||
edge_index: Edge connectivity [2, num_edges]
|
||||
design_features: Design parameters [design_dim] (broadcast to all nodes)
|
||||
design_features: Design parameters [hidden] or [num_nodes, hidden]
|
||||
edge_attr: Edge features [num_edges, edge_dim] (optional)
|
||||
|
||||
Returns:
|
||||
Updated node features [num_nodes, out_channels]
|
||||
"""
|
||||
# Broadcast design features to match number of nodes
|
||||
num_nodes = x.size(0)
|
||||
design_broadcast = design_features.unsqueeze(0).expand(num_nodes, -1)
|
||||
|
||||
# Handle different input shapes for design_features
|
||||
if design_features.dim() == 1:
|
||||
# Single design vector [hidden] -> broadcast to all nodes
|
||||
design_broadcast = design_features.unsqueeze(0).expand(num_nodes, -1)
|
||||
elif design_features.dim() == 2 and design_features.size(0) == num_nodes:
|
||||
# Already per-node [num_nodes, hidden]
|
||||
design_broadcast = design_features
|
||||
elif design_features.dim() == 2 and design_features.size(0) == 1:
|
||||
# Single design [1, hidden] -> broadcast
|
||||
design_broadcast = design_features.expand(num_nodes, -1)
|
||||
else:
|
||||
# Fallback: take mean across batch dimension if needed
|
||||
design_broadcast = design_features.mean(dim=0).unsqueeze(0).expand(num_nodes, -1)
|
||||
|
||||
return self.propagate(
|
||||
edge_index,
|
||||
@@ -319,54 +331,53 @@ class ParametricFieldPredictor(nn.Module):
|
||||
- displacement: (optional) Displacement field [num_nodes, 6]
|
||||
"""
|
||||
x, edge_index, edge_attr = data.x, data.edge_index, data.edge_attr
|
||||
batch = data.batch if hasattr(data, 'batch') else torch.zeros(x.size(0), dtype=torch.long, device=x.device)
|
||||
num_nodes = x.size(0)
|
||||
|
||||
# Handle design params shape
|
||||
# Handle design params shape - ensure 2D [batch_size, design_dim]
|
||||
if design_params.dim() == 1:
|
||||
design_params = design_params.unsqueeze(0)
|
||||
|
||||
# Encode design parameters
|
||||
design_encoded = self.design_encoder(design_params) # [batch, hidden]
|
||||
batch_size = design_params.size(0)
|
||||
|
||||
# For single graph, broadcast design to all nodes
|
||||
if design_encoded.size(0) == 1:
|
||||
design_for_nodes = design_encoded.squeeze(0) # [hidden]
|
||||
else:
|
||||
# For batched graphs, get design for each node based on batch assignment
|
||||
design_for_nodes = design_encoded[batch] # [num_nodes, hidden]
|
||||
# Encode design parameters: [batch_size, design_dim] -> [batch_size, hidden]
|
||||
design_encoded = self.design_encoder(design_params)
|
||||
|
||||
# Encode nodes
|
||||
x = self.node_encoder(x) # [num_nodes, hidden]
|
||||
# Encode nodes (shared across all designs)
|
||||
x_encoded = self.node_encoder(x) # [num_nodes, hidden]
|
||||
|
||||
# Encode edges
|
||||
# Encode edges (shared across all designs)
|
||||
if edge_attr is not None:
|
||||
edge_features = self.edge_encoder(edge_attr) # [num_edges, hidden//2]
|
||||
else:
|
||||
edge_features = None
|
||||
|
||||
# Message passing with design conditioning
|
||||
node_embeddings = x
|
||||
for conv, norm, dropout in zip(self.conv_layers, self.layer_norms, self.dropouts):
|
||||
# Use appropriate design features based on batching
|
||||
if design_params.size(0) == 1:
|
||||
design_input = design_for_nodes
|
||||
else:
|
||||
# For batched case, we need to handle per-node design features
|
||||
design_input = design_for_nodes[0] # Simplified - use first
|
||||
# Process each design in the batch
|
||||
all_graph_features = []
|
||||
|
||||
x_new = conv(x, edge_index, design_input, edge_features)
|
||||
x = x + dropout(x_new) # Residual connection
|
||||
x = norm(x)
|
||||
for i in range(batch_size):
|
||||
# Get design for this sample
|
||||
design_i = design_encoded[i] # [hidden]
|
||||
|
||||
# Global pooling
|
||||
x_mean = global_mean_pool(x, batch) # [batch, hidden]
|
||||
x_max = global_max_pool(x, batch) # [batch, hidden]
|
||||
# Reset node features for this sample
|
||||
x = x_encoded.clone()
|
||||
|
||||
# Concatenate pooled features with design encoding
|
||||
if design_encoded.size(0) == 1 and x_mean.size(0) > 1:
|
||||
design_encoded = design_encoded.expand(x_mean.size(0), -1)
|
||||
# Message passing with design conditioning
|
||||
for conv, norm, dropout in zip(self.conv_layers, self.layer_norms, self.dropouts):
|
||||
x_new = conv(x, edge_index, design_i, edge_features)
|
||||
x = x + dropout(x_new) # Residual connection
|
||||
x = norm(x)
|
||||
|
||||
graph_features = torch.cat([x_mean, x_max, design_encoded], dim=-1) # [batch, 3*hidden]
|
||||
# Global pooling for this sample
|
||||
batch_idx = torch.zeros(num_nodes, dtype=torch.long, device=x.device)
|
||||
x_mean = global_mean_pool(x, batch_idx) # [1, hidden]
|
||||
x_max = global_max_pool(x, batch_idx) # [1, hidden]
|
||||
|
||||
# Concatenate pooled + design features
|
||||
graph_feat = torch.cat([x_mean, x_max, design_encoded[i:i+1]], dim=-1) # [1, 3*hidden]
|
||||
all_graph_features.append(graph_feat)
|
||||
|
||||
# Stack all samples
|
||||
graph_features = torch.cat(all_graph_features, dim=0) # [batch_size, 3*hidden]
|
||||
|
||||
# Predict objectives
|
||||
mass = self.mass_head(graph_features).squeeze(-1)
|
||||
@@ -381,9 +392,9 @@ class ParametricFieldPredictor(nn.Module):
|
||||
'max_stress': max_stress
|
||||
}
|
||||
|
||||
# Optionally return displacement field
|
||||
# Optionally return displacement field (uses last processed x)
|
||||
if return_fields:
|
||||
displacement_field = self.field_decoder(node_embeddings) # [num_nodes, 6]
|
||||
displacement_field = self.field_decoder(x) # [num_nodes, 6]
|
||||
results['displacement'] = displacement_field
|
||||
|
||||
return results
|
||||
|
||||
Reference in New Issue
Block a user