Skip to content

Commit

Permalink
It's working
Browse files Browse the repository at this point in the history
  • Loading branch information
@woodRock
woodRock committed Oct 7, 2024
1 parent 0de9ef6 commit fdbfd90
Show file tree
Hide file tree
Showing 8 changed files with 277 additions and 19 deletions.
Binary file added code/siamese/figures/tree_0.pdf
View file
Binary file not shown.
Binary file added code/siamese/figures/tree_1.pdf
View file
Binary file not shown.
Binary file added code/siamese/figures/tree_2.pdf
View file
Binary file not shown.
Binary file added code/siamese/figures/tree_3.pdf
View file
Binary file not shown.
Binary file added code/siamese/figures/tree_4.pdf
View file
Binary file not shown.
27 changes: 11 additions & 16 deletions code/siamese/gp.py
View file
Original file line number Diff line number Diff line change
@@ -1,18 +1,18 @@
import logging
import random
from multiprocessing import Pool
from tqdm import tqdm
import numpy as np
import torch
import torch.nn.functional as F
from deap import algorithms, base, creator, gp, tools
from sklearn.metrics import balanced_accuracy_score
from sklearn.model_selection import train_test_split


def setup_logging(): # Logging output to a file.
logger = logging.getLogger(__name__)
output = f"logs/instance-recognition/results_0.log"
logging.basicConfig(filename=output, level=logging.INFO, filemode='w')
logging.basicConfig(filename=output, level=logging.DEBUG, filemode='w')
return logger


Expand Down Expand Up @@ -58,7 +58,7 @@ def rand101(x):
creator.create("Individual", list, fitness=creator.FitnessMin)

# Define the number of trees per individual
NUM_TREES = 5
NUM_TREES = 10

# Toolbox initialization
toolbox = base.Toolbox()
Expand All @@ -85,8 +85,6 @@ def evalContrastive(individual, data, alpha=0.5):
labels = []

for x1, x2, label in data:
logger.debug(f"x1: {x1.shape}, {x1}")
logger.debug(f"x2: {x2.shape}, {x2}")
# Evaluate both inputs using the same set of trees (Siamese approach)
outputs1 = torch.tensor(np.array([tree(*x1) for tree in trees]), dtype=torch.float32)
outputs2 = torch.tensor(np.array([tree(*x2) for tree in trees]), dtype=torch.float32)
Expand Down Expand Up @@ -146,7 +144,7 @@ def eaSimpleWithElitism(population, toolbox, cxpb, mutpb, ngen, stats=None,
logger.info(logbook.stream)

# Begin the generational process
for gen in range(1, ngen + 1):
for gen in tqdm(range(1, ngen + 1), desc="Training"):
# Select the next generation individuals
offspring = toolbox.select(population, len(population) - elite_size)

Expand Down Expand Up @@ -223,7 +221,7 @@ def loader_to_list(loader):
toolbox.register("evaluate", evalContrastive, data=train_data)

# GP parameters
pop_size = 100
pop_size = 1000
generations = 50
elite_size = 5 # Number of elite individuals to preserve

Expand All @@ -236,12 +234,12 @@ def loader_to_list(loader):
stats.register("min", np.min)
stats.register("max", np.max)

pool = Pool()
toolbox.register("map", pool.map)

# Run GP with elitism
pop, log = eaSimpleWithElitism(pop, toolbox, cxpb=0.5, mutpb=0.2, ngen=generations,
stats=stats, halloffame=hof, verbose=True, elite_size=elite_size)
with Pool() as pool:
toolbox.register("map", pool.map)
# Run GP with elitism
pop, log = eaSimpleWithElitism(pop, toolbox, cxpb=0.5, mutpb=0.2, ngen=generations,
stats=stats, halloffame=hof, verbose=True, elite_size=elite_size,
train_data=train_data, val_data=val_data)

# Evaluate best individual on validation set
best_individual = hof[0]
Expand Down Expand Up @@ -269,9 +267,6 @@ def loader_to_list(loader):
n.attr["label"] = labels[i]

g.draw(f"figures/tree_{tree_idx}.pdf")

pool.close()
pool.join()

return pop, log, hof

Expand Down
13 changes: 12 additions & 1 deletion code/siamese/gp_vector.py
View file
Original file line number Diff line number Diff line change
@@ -1,3 +1,13 @@
"""
References:
1. Bromley, J., Guyon, I., LeCun, Y., Säckinger, E., & Shah, R. (1993).
Signature verification using a" siamese" time delay neural network.
Advances in neural information processing systems, 6.
2. Koza, J. R. (1994).
Genetic programming II: automatic discovery of reusable programs.
"""

import numpy as np
import operator
from deap import algorithms, base, creator, gp, tools
Expand Down Expand Up @@ -224,7 +234,8 @@ def loader_to_list(loader):

# Run GP with elitism
pop, log = eaSimpleWithElitism(pop, toolbox, cxpb=0.5, mutpb=0.2, ngen=generations,
stats=stats, halloffame=hof, verbose=True, elite_size=elite_size)
stats=stats, halloffame=hof, verbose=True, elite_size=elite_size,
train_data=train_data, val_data=val_data)

# Evaluate best individual on validation set
best_individual = hof[0]
Expand Down
256 changes: 254 additions & 2 deletions code/siamese/logs/instance-recognition/results_0.log
View file
Original file line number Diff line number Diff line change
@@ -1,5 +1,257 @@
INFO:numexpr.utils:Note: NumExpr detected 20 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 8.
INFO:numexpr.utils:NumExpr defaulting to 8 threads.
INFO:__main__:gen nevals avg std min max
0 100 2.27237e+18 2.26098e+19 0.453517 2.27237e+20
INFO:__main__:1 47 299469 2.97935e+06 0.447081 2.99436e+07
0 100 2.59756e+11 2.58451e+12 0.416125 2.59753e+13
INFO:__main__:1 69 3.58209 27.6723 0.39964 278.459
INFO:__main__:
Generation 1: Best Fitness = 0.3996
Balanced accuracy: Train: 0.4765
Validation: 0.4933
INFO:__main__:2 54 5.69927 51.9061 0.39964 522.157
INFO:__main__:
Generation 2: Best Fitness = 0.3996
Balanced accuracy: Train: 0.4905
Validation: 0.5124
INFO:__main__:3 53 275.617 2737.89 0.398553 27517.2
INFO:__main__:
Generation 3: Best Fitness = 0.3986
Balanced accuracy: Train: 0.5143
Validation: 0.4903
INFO:__main__:4 45 4.91037 44.6355 0.397594 449.028
INFO:__main__:
Generation 4: Best Fitness = 0.3976
Balanced accuracy: Train: 0.4916
Validation: 0.4703
INFO:__main__:5 58 16.3391 158.363 0.393966 1592.03
INFO:__main__:
Generation 5: Best Fitness = 0.3940
Balanced accuracy: Train: 0.4916
Validation: 0.4655
INFO:__main__:6 63 15.4036 149.198 0.385693 1499.91
INFO:__main__:
Generation 6: Best Fitness = 0.3857
Balanced accuracy: Train: 0.5069
Validation: 0.4790
INFO:__main__:7 60 0.415005 0.0525142 0.365022 0.866009
INFO:__main__:
Generation 7: Best Fitness = 0.3650
Balanced accuracy: Train: 0.6315
Validation: 0.5661
INFO:__main__:8 50 0.407907 0.0508057 0.365022 0.846733
INFO:__main__:
Generation 8: Best Fitness = 0.3650
Balanced accuracy: Train: 0.6279
Validation: 0.5693
INFO:__main__:9 57 0.403932 0.031149 0.357816 0.53713
INFO:__main__:
Generation 9: Best Fitness = 0.3578
Balanced accuracy: Train: 0.6301
Validation: 0.5738
INFO:__main__:10 49 0.392575 0.0334495 0.357816 0.582461
INFO:__main__:
Generation 10: Best Fitness = 0.3578
Balanced accuracy: Train: 0.6311
Validation: 0.5812
INFO:__main__:11 47 0.377658 0.0327319 0.355398 0.582769
INFO:__main__:
Generation 11: Best Fitness = 0.3570
Balanced accuracy: Train: 0.6273
Validation: 0.5780
INFO:__main__:12 58 0.38311 0.0762926 0.342418 0.862446
INFO:__main__:
Generation 12: Best Fitness = 0.3424
Balanced accuracy: Train: 0.6358
Validation: 0.5489
INFO:__main__:13 55 0.372847 0.0565213 0.342418 0.862447
INFO:__main__:
Generation 13: Best Fitness = 0.3424
Balanced accuracy: Train: 0.6333
Validation: 0.5495
INFO:__main__:14 57 2.12421 17.2641 0.342418 173.898
INFO:__main__:
Generation 14: Best Fitness = 0.3424
Balanced accuracy: Train: 0.6307
Validation: 0.5535
INFO:__main__:15 41 0.365341 0.0382545 0.342418 0.53497
INFO:__main__:
Generation 15: Best Fitness = 0.3424
Balanced accuracy: Train: 0.6226
Validation: 0.5478
INFO:__main__:16 53 262.83 2611.48 0.334201 26246.8
INFO:__main__:
Generation 16: Best Fitness = 0.3342
Balanced accuracy: Train: 0.6524
Validation: 0.5671
INFO:__main__:17 58 0.375734 0.0800477 0.328887 0.844901
INFO:__main__:
Generation 17: Best Fitness = 0.3289
Balanced accuracy: Train: 0.6647
Validation: 0.5729
INFO:__main__:18 63 0.370074 0.0915829 0.325604 0.884079
INFO:__main__:
Generation 18: Best Fitness = 0.3256
Balanced accuracy: Train: 0.6615
Validation: 0.5594
INFO:__main__:19 53 409.831 4074.28 0.322946 40948.4
INFO:__main__:
Generation 19: Best Fitness = 0.3229
Balanced accuracy: Train: 0.6870
Validation: 0.5498
INFO:__main__:20 61 409.907 4074.41 0.3225 40949.7
INFO:__main__:
Generation 20: Best Fitness = 0.3225
Balanced accuracy: Train: 0.6879
Validation: 0.5498
INFO:__main__:21 56 143869 1.43011e+06 0.3225 1.43732e+07
INFO:__main__:
Generation 21: Best Fitness = 0.3225
Balanced accuracy: Train: 0.6879
Validation: 0.5498
INFO:__main__:22 59 0.365721 0.105179 0.322495 1.13313
INFO:__main__:
Generation 22: Best Fitness = 0.3225
Balanced accuracy: Train: 0.6879
Validation: 0.5498
INFO:__main__:23 58 83.4868 827.218 0.320085 8314.2
INFO:__main__:
Generation 23: Best Fitness = 0.3201
Balanced accuracy: Train: 0.6592
Validation: 0.5752
INFO:__main__:24 60 0.350391 0.0653675 0.320083 0.815685
INFO:__main__:
Generation 24: Best Fitness = 0.3201
Balanced accuracy: Train: 0.6606
Validation: 0.5714
INFO:__main__:25 49 0.34648 0.0779952 0.320083 0.966893
INFO:__main__:
Generation 25: Best Fitness = 0.3201
Balanced accuracy: Train: 0.6633
Validation: 0.5681
INFO:__main__:26 67 0.377503 0.0978743 0.320083 0.846316
INFO:__main__:
Generation 26: Best Fitness = 0.3201
Balanced accuracy: Train: 0.6614
Validation: 0.5812
INFO:__main__:27 57 0.34718 0.0642899 0.320083 0.871099
INFO:__main__:
Generation 27: Best Fitness = 0.3201
Balanced accuracy: Train: 0.6587
Validation: 0.5705
INFO:__main__:28 54 0.345693 0.0737162 0.320083 0.830822
INFO:__main__:
Generation 28: Best Fitness = 0.3201
Balanced accuracy: Train: 0.6695
Validation: 0.5686
INFO:__main__:29 58 1.81851 14.5684 0.320083 146.771
INFO:__main__:
Generation 29: Best Fitness = 0.3201
Balanced accuracy: Train: 0.6626
Validation: 0.5736
INFO:__main__:30 63 0.350058 0.0469747 0.320083 0.553101
INFO:__main__:
Generation 30: Best Fitness = 0.3201
Balanced accuracy: Train: 0.6667
Validation: 0.5698
INFO:__main__:31 54 0.354662 0.0925656 0.320083 0.893125
INFO:__main__:
Generation 31: Best Fitness = 0.3201
Balanced accuracy: Train: 0.6559
Validation: 0.5760
INFO:__main__:32 54 4.21969 38.5017 0.320083 387.306
INFO:__main__:
Generation 32: Best Fitness = 0.3201
Balanced accuracy: Train: 0.6602
Validation: 0.5668
INFO:__main__:33 61 57.9642 572.636 0.320083 5755.62
INFO:__main__:
Generation 33: Best Fitness = 0.3201
Balanced accuracy: Train: 0.6646
Validation: 0.5702
INFO:__main__:34 53 0.349023 0.0621259 0.320083 0.830072
INFO:__main__:
Generation 34: Best Fitness = 0.3201
Balanced accuracy: Train: 0.6670
Validation: 0.5672
INFO:__main__:35 49 4.6022 42.4012 0.319496 426.488
INFO:__main__:
Generation 35: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6532
Validation: 0.5671
INFO:__main__:36 54 0.362307 0.10534 0.319496 0.945707
INFO:__main__:
Generation 36: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6608
Validation: 0.5669
INFO:__main__:37 60 100.332 994.838 0.319496 9998.84
INFO:__main__:
Generation 37: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6615
Validation: 0.5665
INFO:__main__:38 52 0.351276 0.0776125 0.319496 0.836795
INFO:__main__:
Generation 38: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6640
Validation: 0.5738
INFO:__main__:39 53 130.174 1291.75 0.319496 12982.9
INFO:__main__:
Generation 39: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6597
Validation: 0.5717
INFO:__main__:40 54 0.352706 0.0824096 0.319496 0.877868
INFO:__main__:
Generation 40: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6636
Validation: 0.5634
INFO:__main__:41 51 0.455819 1.00189 0.319496 10.3892
INFO:__main__:
Generation 41: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6609
Validation: 0.5692
INFO:__main__:42 55 0.354377 0.0924357 0.319496 0.878259
INFO:__main__:
Generation 42: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6649
Validation: 0.5695
INFO:__main__:43 51 0.348531 0.0795894 0.319496 0.878316
INFO:__main__:
Generation 43: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6613
Validation: 0.5718
INFO:__main__:44 53 0.341779 0.0586394 0.319496 0.841007
INFO:__main__:
Generation 44: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6537
Validation: 0.5701
INFO:__main__:45 52 16.4231 159.861 0.319496 1607.02
INFO:__main__:
Generation 45: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6632
Validation: 0.5721
INFO:__main__:46 59 0.362323 0.0957811 0.319496 0.878843
INFO:__main__:
Generation 46: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6651
Validation: 0.5628
INFO:__main__:47 56 0.356455 0.0921035 0.319496 0.848626
INFO:__main__:
Generation 47: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6627
Validation: 0.5643
INFO:__main__:48 63 0.395166 0.422891 0.319496 4.51286
INFO:__main__:
Generation 48: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6605
Validation: 0.5675
INFO:__main__:49 67 12179.7 121100 0.319496 1.2171e+06
INFO:__main__:
Generation 49: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6596
Validation: 0.5747
INFO:__main__:50 46 0.348531 0.0672949 0.319496 0.878906
INFO:__main__:
Generation 50: Best Fitness = 0.3195
Balanced accuracy: Train: 0.6609
Validation: 0.5744
INFO:__main__:Best individual fitness on validation set: 0.3666401320401358
INFO:__main__:Balanced Accuracy Score of the best individual on validation set: 0.5718
INFO:__main__:Printing the GP trees

0 comments on commit fdbfd90

Please sign in to comment.