File size: 15,233 Bytes
c536443 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 |
"""Analysis utilities for neural networks.
This module provides functions for analyzing neural network architectures,
including complexity measures and structural properties.
"""
import numpy as np
import networkx as nx
import matplotlib.pyplot as plt
from typing import Dict, Tuple, Union, Optional, List, Any
from .network import Network
from .genome import Genome
from collections import defaultdict
import os
def analyze_network_complexity(network: Network) -> Dict[str, Any]:
"""Analyze the complexity of a neural network.
Computes various complexity metrics including:
1. Number of nodes by type (input, hidden, output)
2. Number of connections
3. Network density
4. Activation functions used
Args:
network: Network instance to analyze
Returns:
Dictionary containing complexity metrics
"""
# Get network structure
genome = network.genome
# Count nodes by type
n_input = genome.input_size
n_hidden = len(genome.hidden_nodes)
n_output = genome.output_size
# Count connections
n_connections = len(genome.connections)
# Calculate connectivity density
n_possible = (n_input + n_hidden + n_output) * (n_hidden + n_output) # No connections to input
density = n_connections / n_possible if n_possible > 0 else 0
# Get activation functions (currently only ReLU)
activation_functions = {'relu': n_hidden + n_output} # All nodes use ReLU
return {
'n_input': n_input,
'n_hidden': n_hidden,
'n_output': n_output,
'n_connections': n_connections,
'density': density,
'activation_functions': activation_functions
}
def get_network_stats(network: Network) -> Dict[str, float]:
"""Get statistical measures of network properties.
Computes various statistics about the network structure and parameters:
- Number of nodes and connections
- Average and std of weights and biases
- Network density and depth
Args:
network: Network instance to analyze
Returns:
Dictionary containing network statistics
"""
stats = {}
# Node counts
stats['n_nodes'] = network.n_nodes
stats['n_hidden'] = network.n_nodes - network.input_size - network.output_size
# Connection stats
weights = np.array(list(network.weights.values()))
stats['n_connections'] = len(weights)
stats['weight_mean'] = float(np.mean(weights))
stats['weight_std'] = float(np.std(weights))
# Bias stats
biases = np.array(list(network.bias.values()))
stats['n_biases'] = len(biases)
stats['bias_mean'] = float(np.mean(biases))
stats['bias_std'] = float(np.std(biases))
# Connectivity
n_possible = network.n_nodes * (network.n_nodes - 1)
stats['density'] = len(weights) / n_possible if n_possible > 0 else 0
# Compute approximate network depth
weight_matrix = network.weight_matrix
depth = 0
visited = set(range(network.input_size))
frontier = visited.copy()
while frontier and depth < network.n_nodes:
next_frontier = set()
for node in frontier:
for next_node in range(network.n_nodes):
if weight_matrix[node, next_node] != 0 and next_node not in visited:
next_frontier.add(next_node)
visited.add(next_node)
frontier = next_frontier
if frontier:
depth += 1
stats['depth'] = depth
return stats
def visualize_network_architecture(network: Network, save_path: Optional[str] = None):
"""Visualize the network architecture using networkx.
Creates a layered visualization of the neural network with:
- Input nodes in red (leftmost layer)
- Hidden nodes in blue (middle layer)
- Output nodes in green (rightmost layer)
- Connections shown as arrows with thickness proportional to weight
Args:
network: Network instance to visualize
save_path: Optional path to save the visualization
Returns:
matplotlib figure object or None if visualization fails
"""
try:
import networkx as nx
import matplotlib.pyplot as plt
genome = network.genome
G = nx.DiGraph()
# Calculate layout parameters
n_inputs = len([node for node in genome.node_genes.values() if node.node_type == 'input'])
n_outputs = len([node for node in genome.node_genes.values() if node.node_type == 'output'])
hidden_nodes = [node.node_id for node in genome.node_genes.values() if node.node_type == 'hidden']
n_hidden = len(hidden_nodes)
# Layout parameters
node_spacing = 1.0 # Vertical spacing between nodes in same layer
layer_spacing = 2.0 # Horizontal spacing between layers
# Initialize position and color dictionaries
pos = {}
node_colors = {}
# Add input nodes (leftmost layer)
input_start_y = -(n_inputs - 1) * node_spacing / 2 # Center vertically
input_nodes = [node.node_id for node in genome.node_genes.values() if node.node_type == 'input']
for i, node_idx in enumerate(input_nodes):
pos[node_idx] = (0, input_start_y + i * node_spacing)
node_colors[node_idx] = 'lightcoral' # Light red for input nodes
# Add hidden nodes (middle layer)
if hidden_nodes:
hidden_start_y = -(n_hidden - 1) * node_spacing / 2 # Center vertically
for i, node_idx in enumerate(hidden_nodes):
pos[node_idx] = (layer_spacing, hidden_start_y + i * node_spacing)
node_colors[node_idx] = 'lightblue' # Light blue for hidden nodes
# Add output nodes (rightmost layer)
output_start_y = -(n_outputs - 1) * node_spacing / 2 # Center vertically
output_nodes = [node.node_id for node in genome.node_genes.values() if node.node_type == 'output']
for i, node_idx in enumerate(output_nodes):
pos[node_idx] = (2 * layer_spacing, output_start_y + i * node_spacing)
node_colors[node_idx] = 'lightgreen' # Light green for output nodes
# Add bias node if present
bias_node = [node.node_id for node in genome.node_genes.values() if node.node_type == 'bias']
if bias_node:
pos[bias_node[0]] = (0, input_start_y - node_spacing) # Place below input nodes
node_colors[bias_node[0]] = 'yellow' # Yellow for bias node
# Add all nodes to graph and ensure they have colors and positions
for node_id in genome.node_genes:
G.add_node(node_id)
if node_id not in node_colors: # Assign default color if not already assigned
node_type = genome.node_genes[node_id].node_type
if node_type == 'input':
node_colors[node_id] = 'lightcoral'
elif node_type == 'hidden':
node_colors[node_id] = 'lightblue'
elif node_type == 'output':
node_colors[node_id] = 'lightgreen'
elif node_type == 'bias':
node_colors[node_id] = 'yellow'
else:
node_colors[node_id] = 'gray' # Default color for unknown types
# Ensure node has a position
if node_id not in pos:
# Place unknown nodes in middle layer
pos[node_id] = (layer_spacing, 0)
# Add connections
for conn in genome.connection_genes:
if conn.enabled:
# Scale connection width by weight
width = abs(conn.weight) * 2.0
# Use red for negative weights, green for positive
color = 'red' if conn.weight < 0 else 'green'
alpha = min(abs(conn.weight), 1.0) # Transparency based on weight magnitude
G.add_edge(conn.source, conn.target, weight=width, color=color, alpha=alpha)
# Set up the plot
fig = plt.figure(figsize=(12, 8))
# Draw nodes with colors
nx.draw_networkx_nodes(G, pos, node_color=[node_colors[node] for node in G.nodes()],
node_size=800, alpha=0.8)
# Draw edges with width proportional to weight
edge_weights = [G.get_edge_data(edge[0], edge[1])['weight'] for edge in G.edges()]
if edge_weights: # Only draw edges if there are any
max_weight = max(edge_weights)
normalized_weights = [3 * w / max_weight for w in edge_weights] # Scale for visibility
nx.draw_networkx_edges(G, pos, edge_color=[G.get_edge_data(edge[0], edge[1])['color'] for edge in G.edges()],
width=normalized_weights,
alpha=[G.get_edge_data(edge[0], edge[1])['alpha'] for edge in G.edges()],
arrows=True, arrowsize=20)
# Add node labels
nx.draw_networkx_labels(G, pos, font_size=10)
plt.title("Neural Network Architecture")
plt.axis('off') # Hide axes
if save_path:
# Ensure the directory exists
os.makedirs(os.path.dirname(save_path), exist_ok=True)
plt.savefig(save_path, bbox_inches='tight', dpi=300)
plt.close(fig) # Close the figure to free memory
return fig
except Exception as e:
print(f"Error visualizing network: {str(e)}")
return None
def plot_activation_distribution(population: List[Genome], save_path: Optional[str] = None):
"""Plot the distribution of node types in the population.
Args:
population: List of genomes in the population
save_path: Optional path to save the plot
Returns:
matplotlib figure object or None if plotting fails
"""
try:
# Count nodes by type for each genome
node_type_counts = defaultdict(int)
for genome in population:
node_type_counts['input'] += genome.input_size
node_type_counts['hidden'] += len(genome.hidden_nodes)
node_type_counts['output'] += genome.output_size
if not node_type_counts:
print("No nodes found in population")
return None
# Create bar plot
fig = plt.figure(figsize=(10, 6))
# Get node types and counts
node_types = list(node_type_counts.keys())
counts = list(node_type_counts.values())
# Create bars with different colors
colors = {'input': 'lightcoral', 'hidden': 'lightblue', 'output': 'lightgreen'}
plt.bar(node_types, counts, color=[colors[t] for t in node_types], alpha=0.7)
# Customize plot
plt.title('Distribution of Node Types in Population')
plt.xlabel('Node Type')
plt.ylabel('Total Count')
# Add count labels on top of bars
for i, count in enumerate(counts):
plt.text(i, count, str(count), ha='center', va='bottom')
plt.tight_layout()
# Save or display
if save_path:
# Ensure the directory exists
os.makedirs(os.path.dirname(save_path), exist_ok=True)
plt.savefig(save_path, bbox_inches='tight', dpi=300)
plt.close(fig) # Close the figure to free memory
return fig
except Exception as e:
print(f"Error plotting activation distribution: {str(e)}")
return None
def analyze_evolution_trends(stats: Dict, save_dir: str) -> None:
"""Analyze and plot evolution trends from training history.
Args:
stats: Dictionary containing training statistics
save_dir: Directory to save plots
"""
try:
# Create plots directory if it doesn't exist
os.makedirs(save_dir, exist_ok=True)
# Check if we have any stats to plot
if not stats or 'mean_fitness' not in stats or not stats['mean_fitness']:
print("No evolution stats available yet")
return
# Extract metrics over generations
generations = list(range(len(stats['mean_fitness'])))
if not generations: # No data points yet
print("No generations completed yet")
return
metrics = {
'Fitness': {
'mean': stats.get('mean_fitness', []),
'best': stats.get('best_fitness', [])
},
'Complexity': {
'mean': stats.get('mean_complexity', []),
'best': stats.get('best_complexity', [])
}
}
# Plot each metric
for metric_name, metric_data in metrics.items():
# Verify we have data for this metric
if not metric_data['mean'] or not metric_data['best']:
print(f"No data available for {metric_name}")
continue
# Verify data lengths match
if len(generations) != len(metric_data['mean']) or len(generations) != len(metric_data['best']):
print(f"Data length mismatch for {metric_name}")
continue
fig = plt.figure(figsize=(10, 6))
# Plot mean and best values
plt.plot(generations, metric_data['mean'], label=f'Mean {metric_name}', alpha=0.7)
plt.plot(generations, metric_data['best'], label=f'Best {metric_name}', alpha=0.7)
plt.title(f'{metric_name} Over Generations')
plt.xlabel('Generation')
plt.ylabel(metric_name)
plt.legend()
plt.grid(True, alpha=0.3)
# Save plot
save_path = os.path.join(save_dir, f'{metric_name.lower()}_trends.png')
plt.savefig(save_path, bbox_inches='tight', dpi=300)
plt.close(fig) # Close the figure to free memory
# Plot species counts if available
if 'n_species' in stats and stats['n_species']:
n_species = stats['n_species']
if len(generations) == len(n_species): # Verify data length matches
fig = plt.figure(figsize=(10, 6))
plt.plot(generations, n_species, label='Number of Species', alpha=0.7)
plt.title('Number of Species Over Generations')
plt.xlabel('Generation')
plt.ylabel('Number of Species')
plt.legend()
plt.grid(True, alpha=0.3)
save_path = os.path.join(save_dir, 'species_trends.png')
plt.savefig(save_path, bbox_inches='tight', dpi=300)
plt.close(fig) # Close the figure to free memory
else:
print("Species count data length mismatch")
except Exception as e:
print(f"Error analyzing evolution trends: {str(e)}")
|