Spaces:
Sleeping
Sleeping
File size: 18,032 Bytes
5e35da7 b930aac 5e35da7 78d4100 fa0da9f 78d4100 5e35da7 9d25e02 fa0da9f 5e35da7 fa0da9f 5e35da7 78d4100 b930aac 5e35da7 b930aac 5e35da7 b930aac 5e35da7 b930aac fa0da9f b930aac 5e35da7 78d4100 5e35da7 78d4100 5e35da7 78d4100 5e35da7 b930aac 5e35da7 78d4100 b930aac 78d4100 5e35da7 78d4100 5e35da7 78d4100 5e35da7 78d4100 b930aac 5e35da7 78d4100 5e35da7 78d4100 fdda688 b930aac 78d4100 b930aac 78d4100 b930aac 78d4100 5e35da7 78d4100 5e35da7 fa0da9f 5e35da7 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 |
import random
import numpy as np
import streamlit as st
from concurrent.futures import ThreadPoolExecutor
from functools import lru_cache
# import all functions from src.backend.chatbot
from src.backend.chatbot import *
def genetic_algorithm_plants(model, demo_lite):
# Define the compatibility matrix
compatibility_matrix = st.session_state.full_mat
# Define the list of plants
plant_list = st.session_state.plant_list
# Define the user-selected plants, number of plant beds, and constraints
user_plants = st.session_state.input_plants_raw
num_plant_beds = st.session_state.n_plant_beds
# 1 <= min_species_per_bed <= max_species_per_bed <= len(user_plants)
min_species_per_bed = st.session_state.min_species
# max_species_per_bed >= floor(length(user_plants)-(min_species_per_bed*num_plant_beds-1) & max_species_per_bed <= len(user_plants)
max_species_per_bed = st.session_state.max_species
# Genetic Algorithm parameters
population_size = st.session_state.population_size
num_generations = st.session_state.num_generations
tournament_size = st.session_state.tournament_size
crossover_rate = st.session_state.crossover_rate
mutation_rate = st.session_state.mutation_rate
seed_population_rate = st.session_state.seed_population_rate
# OPTIMIZATION: Create plant name to index mapping for O(1) lookups
plant_to_index = {plant: idx for idx, plant in enumerate(plant_list)}
# OPTIMIZATION: Convert compatibility matrix to numpy array ONCE (not in loop)
compat_array = np.array(compatibility_matrix)
# OPTIMIZATION: Fitness cache to avoid recalculating fitness for the same grouping
fitness_cache = {}
def generate_initial_population(model, demo_lite):
population = []
# OPTIMIZATION: Only call LLM if seed_population_rate > 0 and not demo_lite
# This was a MAJOR bottleneck - LLM initialization and inference is very slow
num_seeds = int(population_size * st.session_state.seed_population_rate)
if num_seeds > 0 and not demo_lite and model is not None:
# we generate just one seed grouping for this beta language model suggestion feature
seed_grouping = get_language_model_suggestions(model, demo_lite)
if seed_grouping != "no response yet":
valid_seed_grouping = validate_and_replace(seed_grouping)
population.append(valid_seed_grouping)
print(f" Added 1 LLM-generated seed to population")
# Fill the rest of the population with random groupings
# IMPORTANT: Validate all initial groupings for best solution quality
# A good initial population is critical for finding optimal solutions
while len(population) < population_size:
random_grouping = generate_random_grouping()
# Always validate to ensure we start with high-quality individuals
valid_grouping = validate_and_replace(random_grouping)
population.append(valid_grouping)
return population
def generate_random_grouping():
random.shuffle(user_plants)
remaining_plants = user_plants.copy()
grouping = []
total_plants = len(user_plants)
plants_per_bed = total_plants // num_plant_beds
extra_plants = total_plants % num_plant_beds
for bed_index in range(num_plant_beds):
if bed_index < extra_plants:
# Distribute extra plants among the first few beds
num_species_in_bed = plants_per_bed + 1
else:
num_species_in_bed = plants_per_bed
# Ensure the bed size is within the min and max constraints
num_species_in_bed = max(
min_species_per_bed, min(num_species_in_bed, max_species_per_bed)
)
bed = remaining_plants[:num_species_in_bed]
remaining_plants = remaining_plants[num_species_in_bed:]
grouping.append(bed)
return grouping
# Perform crossover between two parents, preserving at least one occurrence of each plant
def crossover(parent1, parent2):
if random.random() < crossover_rate:
crossover_point = random.randint(1, num_plant_beds - 1)
child1 = parent1[:crossover_point] + parent2[crossover_point:]
child2 = parent2[:crossover_point] + parent1[crossover_point:]
# Ensure each plant appears at least once in the offspring
for plant in user_plants:
if all(plant not in bed for bed in child1):
# Find a bed with fewer species and add the missing plant
min_bed_index = min(
range(len(child1)), key=lambda i: len(child1[i])
)
child1[min_bed_index].append(plant)
if all(plant not in bed for bed in child2):
# Find a bed with fewer species and add the missing plant
min_bed_index = min(
range(len(child2)), key=lambda i: len(child2[i])
)
child2[min_bed_index].append(plant)
return child1, child2
else:
return parent1, parent2
# Perform mutation on an individual, ensuring no bed exceeds the maximum species constraint
def mutate(individual):
if random.random() < mutation_rate:
mutated_bed = random.randint(0, num_plant_beds - 1)
species_in_bed = individual[mutated_bed]
# Remove excess species if there are more than the maximum constraint
if len(species_in_bed) > max_species_per_bed:
species_in_bed = random.sample(species_in_bed, max_species_per_bed)
# Add missing plants by performing swaps between current species and missing plants
missing_plants = [
plant for plant in user_plants if plant not in species_in_bed
]
num_missing_plants = min(
len(missing_plants), max_species_per_bed - len(species_in_bed)
)
for _ in range(num_missing_plants):
swap_species = random.choice(missing_plants)
missing_plants.remove(swap_species)
species_in_bed.append(swap_species)
species_in_bed.remove(random.choice(species_in_bed))
individual[mutated_bed] = species_in_bed
return individual
# calculate the fitness score of the grouping
def calculate_fitness(grouping):
# OPTIMIZATION: Create a hashable key for caching
grouping_key = tuple(tuple(sorted(bed)) for bed in grouping)
if grouping_key in fitness_cache:
return fitness_cache[grouping_key]
positive_reward_factor = 1000
negative_penalty_factor = 2000
# define penalties for not meeting constraints
penalty_for_exceeding_max = 500
penalty_for_not_meeting_min = 500
penalty_for_not_having_all_plants = 1000
score = 0
# VECTORIZED FITNESS CALCULATION - Much faster with numpy
for bed in grouping:
if len(bed) < 2:
continue
# Convert plant names to indices in bulk
bed_indices = np.array([plant_to_index[plant] for plant in bed])
# Get all pairwise compatibility scores using numpy advanced indexing
# This avoids nested loops and is 10-100x faster
n = len(bed_indices)
i_indices, j_indices = np.triu_indices(n, k=1)
# Vectorized compatibility score extraction (using pre-converted array)
compat_scores = compat_array[bed_indices[i_indices], bed_indices[j_indices]]
# Vectorized reward/penalty calculation
positive_scores = compat_scores[compat_scores > 0].sum() * positive_reward_factor
negative_scores = compat_scores[compat_scores < 0].sum() * negative_penalty_factor
score += positive_scores + negative_scores
# apply penalties for not meeting constraints (vectorized)
bed_sizes = np.array([len(bed) for bed in grouping])
score -= np.sum(bed_sizes > max_species_per_bed) * penalty_for_exceeding_max
score -= np.sum(bed_sizes < min_species_per_bed) * penalty_for_not_meeting_min
if len(set(plant for bed in grouping for plant in bed)) < len(user_plants):
score -= penalty_for_not_having_all_plants
# OPTIMIZATION: Cache the result
fitness_cache[grouping_key] = score
return score
# Perform tournament selection
def tournament_selection(population, population_fitness):
# OPTIMIZATION: Use pre-calculated fitness scores
selected = []
for _ in range(population_size):
participants_idx = random.sample(range(len(population)), tournament_size)
winner_idx = max(participants_idx, key=lambda idx: population_fitness[idx])
selected.append(population[winner_idx])
return selected
# OPTIMIZATION: Parallel fitness calculation for speed
def calculate_fitness_parallel(individuals):
"""Calculate fitness for multiple individuals in parallel"""
if len(individuals) <= 10:
# For small populations, parallel overhead isn't worth it
return [calculate_fitness(ind) for ind in individuals]
# Use ThreadPoolExecutor for parallel computation
with ThreadPoolExecutor(max_workers=4) as executor:
return list(executor.map(calculate_fitness, individuals))
# Perform replacement of the population with the offspring, ensuring maximum species constraint is met
def replacement(population, offspring, population_fitness):
# OPTIMIZATION: Use pre-calculated fitness and avoid re-sorting
# Calculate fitness for offspring in parallel
offspring_fitness = calculate_fitness_parallel(offspring)
# Adjust the offspring to meet the maximum species constraint
adjusted_offspring = []
adjusted_fitness = []
for idx, individual in enumerate(offspring):
for bed_idx in range(num_plant_beds):
species_in_bed = individual[bed_idx]
if len(species_in_bed) > max_species_per_bed:
species_in_bed = random.sample(species_in_bed, max_species_per_bed)
individual[bed_idx] = species_in_bed
adjusted_offspring.append(individual)
adjusted_fitness.append(offspring_fitness[idx])
# Combine population and offspring with their fitness scores
combined = list(zip(population + adjusted_offspring, population_fitness + adjusted_fitness))
# Sort by fitness and take top population_size individuals
combined.sort(key=lambda x: x[1], reverse=True)
new_population = [ind for ind, _ in combined[:population_size]]
new_fitness = [fit for _, fit in combined[:population_size]]
return new_population, new_fitness
# Genetic Algorithm main function
def genetic_algorithm(model, demo_lite):
population = generate_initial_population(model, demo_lite)
# OPTIMIZATION: Calculate fitness once for initial population (in parallel)
population_fitness = calculate_fitness_parallel(population)
for generation in range(num_generations):
print(f"Generation {generation + 1}")
selected_population = tournament_selection(population, population_fitness)
offspring = []
for _ in range(population_size // 2):
parent1 = random.choice(selected_population)
parent2 = random.choice(selected_population)
child1, child2 = crossover(parent1, parent2)
child1 = mutate(child1)
child2 = mutate(child2)
offspring.extend([child1, child2])
# OPTIMIZATION: Pass fitness and get updated fitness back
population, population_fitness = replacement(population, offspring, population_fitness)
# OPTIMIZATION: Validate periodically to ensure quality (every 10 generations)
# Too frequent = slow, too rare = poor quality. Every 10 is a good balance.
if generation % 10 == 0 or generation == num_generations - 1:
# Only validate if needed - most individuals are valid
validated_count = 0
invalid_indices = []
# Quick check which individuals need validation
for i in range(len(population)):
plants_in_grouping = set(plant for bed in population[i] for plant in bed)
if len(plants_in_grouping) != len(user_plants):
invalid_indices.append(i)
# Parallel validation for invalid individuals
if invalid_indices:
invalid_individuals = [population[i] for i in invalid_indices]
validated_individuals = [validate_and_replace(ind) for ind in invalid_individuals]
# Update population and recalculate fitness
for idx, validated_ind in zip(invalid_indices, validated_individuals):
population[idx] = validated_ind
population_fitness[idx] = calculate_fitness(validated_ind)
validated_count = len(invalid_indices)
if validated_count > 0:
print(f" Validated {validated_count} individuals")
# Find best solution
best_idx = max(range(len(population)), key=lambda i: population_fitness[i])
best_grouping = population[best_idx]
# Final validation of best solution only
best_grouping = validate_and_replace(best_grouping)
best_fitness = calculate_fitness(best_grouping)
print(f"Best Grouping: {best_grouping}")
print(f"Fitness Score: {best_fitness}")
st.session_state.best_grouping = best_grouping
st.session_state.best_fitness = best_fitness
# st.write(f"Best Grouping: {best_grouping}")
# st.write(f"Fitness Score: {best_fitness}")
return best_grouping
# def validate_and_replace(grouping):
# print("Grouping structure before validation:", grouping)
# all_plants = set(user_plants)
# for bed in grouping:
# all_plants -= set(bed)
# # Replace missing plants
# for missing_plant in all_plants:
# replaced = False
# for bed in grouping:
# if len(set(bed)) != len(bed): # Check for duplicates
# for i, plant in enumerate(bed):
# if bed.count(plant) > 1: # Found a duplicate
# bed[i] = missing_plant
# replaced = True
# break
# if replaced:
# break
# # If no duplicates were found, replace a random plant
# if not replaced:
# random_bed = random.choice(grouping)
# random_bed[random.randint(0, len(random_bed) - 1)] = missing_plant
# return grouping
############
############ experimental
def adjust_grouping(grouping):
# Determine the plants that are missing in the grouping
plants_in_grouping = set(plant for bed in grouping for plant in bed)
missing_plants = set(user_plants) - plants_in_grouping
for missing_plant in missing_plants:
# Find a bed that can accommodate the missing plant without exceeding max_species_per_bed
suitable_bed = next(
(bed for bed in grouping if len(bed) < max_species_per_bed), None
)
if suitable_bed is not None:
suitable_bed.append(missing_plant)
else:
# If no suitable bed is found, replace a random plant in a random bed
random_bed = random.choice(grouping)
random_bed[random.randint(0, len(random_bed) - 1)] = missing_plant
# Ensure min_species_per_bed and max_species_per_bed constraints
for bed in grouping:
while len(bed) < min_species_per_bed:
additional_plant = random.choice(
[plant for plant in user_plants if plant not in bed]
)
bed.append(additional_plant)
while len(bed) > max_species_per_bed:
bed.remove(random.choice(bed))
return grouping
def validate_and_replace(grouping):
# Try multiple configurations to find the best valid grouping
# This is important for solution quality - don't skip this!
best_grouping = None
best_fitness = float("-inf")
# Try 3 different configurations (balanced between speed and quality)
for _ in range(3):
temp_grouping = [bed.copy() for bed in grouping]
temp_grouping = adjust_grouping(temp_grouping)
current_fitness = calculate_fitness(temp_grouping)
if current_fitness > best_fitness:
best_fitness = current_fitness
best_grouping = temp_grouping
return best_grouping
############
def get_language_model_suggestions(model, demo_lite):
# This returns a list of seed groupings based on the compatibility matrix
st.session_state.seed_groupings = get_seed_groupings_from_LLM(model, demo_lite)
return st.session_state.seed_groupings
# Run the genetic algorithm
best_grouping = genetic_algorithm(model, demo_lite)
return best_grouping
|