Spaces:
Sleeping
Sleeping
File size: 62,707 Bytes
67fb03c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 |
"""
PhysicsNeMo-Style Dataset Analysis
Following NVIDIA's PhysicsNeMo-Curator methodology for external aerodynamics analysis.
This module provides a structured approach to analyzing datasets with UMAP visualization.
"""
import os
import yaml
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from pathlib import Path
from typing import List, Dict, Tuple, Optional, Any
import multiprocessing as mp
from tqdm import tqdm
from dataclasses import dataclass
from abc import ABC, abstractmethod
# VTK imports
try:
import vtk
from vtk.util.numpy_support import vtk_to_numpy
except ImportError:
print("VTK not found. Please install it with: pip install vtk")
exit(1)
# UMAP import
try:
import umap
except ImportError:
print("UMAP not found. Please install it with: pip install umap-learn")
exit(1)
# Scikit-learn for preprocessing
from sklearn.preprocessing import StandardScaler, RobustScaler
from sklearn.decomposition import PCA
from sklearn.manifold import TSNE
from sklearn.neighbors import NearestNeighbors
# Clustering
try:
import cuml
from cuml.cluster import HDBSCAN
CUML_AVAILABLE = True
except ImportError:
try:
import hdbscan
from hdbscan import prediction
HDBSCAN_AVAILABLE = True
CUML_AVAILABLE = False
except ImportError:
HDBSCAN_AVAILABLE = False
CUML_AVAILABLE = False
@dataclass
class DataFolder:
"""Configuration for a single data folder."""
path: str
label: str
color: str = "blue"
train_path: str = ""
test_path: str = ""
@dataclass
class AnalysisConfig:
"""Configuration class for the analysis pipeline."""
data_folders: List[DataFolder]
file_pattern: str = "*.vtp"
max_files_per_folder: int = -1
use_multiprocessing: bool = True
num_workers: int = 4
# Feature extraction
include_pressure_stats: bool = False
max_points_for_analysis: int = 50000
sampling_method: str = "uniform"
subsampling_distance: float = 0.1
normalize_features: bool = True
scaling_method: str = "robust"
feature_selection: Dict[str, bool] = None
# UMAP parameters
n_components: int = 2
n_neighbors: int = 15
min_dist: float = 0.1
metric: str = "euclidean"
random_state: int = 42
n_epochs: int = 1000
learning_rate: float = 1.0
# Clustering parameters
clustering_algorithm: str = "hdbscan"
min_cluster_size: int = 10
cluster_selection_epsilon: float = 1.5
allow_single_cluster: bool = True
dbscan_eps: float = 0.5
dbscan_min_samples: int = 3
# Visualization
figsize: Tuple[int, int] = (12, 10)
point_size: int = 50
alpha: float = 0.7
colormap: str = "viridis"
show_labels: bool = False
save_plot: bool = True
output_path: str = "car_umap_visualization.png"
dpi: int = 300
# Output
save_embedding: bool = True
save_features: bool = True
save_labels: bool = True
@classmethod
def from_yaml(cls, config_path: str) -> 'AnalysisConfig':
"""Load configuration from YAML file."""
with open(config_path, 'r') as f:
config_data = yaml.safe_load(f)
# Extract data section
data = config_data.get('data', {})
features = config_data.get('features', {})
umap_config = config_data.get('umap', {})
clustering = config_data.get('clustering', {})
visualization = config_data.get('visualization', {})
output = config_data.get('output', {})
# Parse data folders
data_folders_list = data.get('data_folders', [])
if not data_folders_list:
# Fallback to single data folder for backward compatibility
single_folder = data.get('data_folder', '')
if single_folder:
data_folders_list = [{'path': single_folder, 'label': 'Dataset', 'color': 'blue', 'train_path': '', 'test_path': ''}]
data_folders = [DataFolder(**folder) for folder in data_folders_list]
return cls(
# Data parameters
data_folders=data_folders,
file_pattern=data.get('file_pattern', '*.vtp'),
max_files_per_folder=data.get('max_files_per_folder', -1),
use_multiprocessing=data.get('use_multiprocessing', True),
num_workers=data.get('num_workers', 4),
# Feature extraction parameters
include_pressure_stats=features.get('include_pressure_stats', False),
max_points_for_analysis=features.get('max_points_for_analysis', 50000),
sampling_method=features.get('sampling_method', 'uniform'),
subsampling_distance=features.get('subsampling_distance', 0.1),
normalize_features=features.get('normalize_features', True),
scaling_method=features.get('scaling_method', 'robust'),
feature_selection=features.get('feature_selection', {}),
# UMAP parameters
n_components=umap_config.get('n_components', 2),
n_neighbors=umap_config.get('n_neighbors', 15),
min_dist=umap_config.get('min_dist', 0.1),
metric=umap_config.get('metric', 'euclidean'),
random_state=umap_config.get('random_state', 42),
n_epochs=umap_config.get('n_epochs', 1000),
learning_rate=umap_config.get('learning_rate', 1.0),
# Clustering parameters
clustering_algorithm=clustering.get('algorithm', 'hdbscan'),
min_cluster_size=clustering.get('min_cluster_size', 10),
cluster_selection_epsilon=clustering.get('cluster_selection_epsilon', 1.5),
allow_single_cluster=clustering.get('allow_single_cluster', True),
dbscan_eps=clustering.get('eps', 0.5),
dbscan_min_samples=clustering.get('min_samples', 3),
# Visualization parameters
figsize=tuple(visualization.get('figsize', [12, 10])),
point_size=visualization.get('point_size', 50),
alpha=visualization.get('alpha', 0.7),
colormap=visualization.get('colormap', 'viridis'),
show_labels=visualization.get('show_labels', False),
save_plot=visualization.get('save_plot', True),
output_path=visualization.get('output_path', 'umap_visualization.png'),
dpi=visualization.get('dpi', 300),
# Output parameters
save_embedding=output.get('save_embedding', True),
save_features=output.get('save_features', True),
save_labels=output.get('save_labels', True)
)
class GeometryProcessor(ABC):
"""Abstract base class for geometry processing."""
@abstractmethod
def process(self, polydata: vtk.vtkPolyData) -> Dict[str, float]:
"""Process geometry and extract features."""
pass
class PointCloudProcessor(GeometryProcessor):
"""Point cloud feature extraction following PhysicsNeMo methodology."""
def __init__(self, config: AnalysisConfig):
self.config = config
def process(self, polydata: vtk.vtkPolyData) -> Dict[str, float]:
"""Extract point cloud features from VTK polydata."""
features = {}
# Get points
points = polydata.GetPoints()
num_points = points.GetNumberOfPoints()
if num_points == 0:
return features
# Convert to numpy array
points_array = vtk_to_numpy(points.GetData())
# Check if this is a mesh (has surface cells) or just a point cloud
has_surface_cells = self._has_surface_cells(polydata)
# If it's a mesh, subsample points from the surface
if has_surface_cells:
points_array = self._subsample_mesh_points(polydata, points_array)
num_points = len(points_array)
# Extract features
all_features = {}
all_features.update(self._extract_basic_features(points_array, num_points, polydata))
all_features.update(self._extract_shape_features(points_array))
# Filter features based on feature selection
if self.config.feature_selection:
features = {name: value for name, value in all_features.items()
if self.config.feature_selection.get(name, True)}
else:
features = all_features
return features
def _has_surface_cells(self, polydata: vtk.vtkPolyData) -> bool:
"""Check if polydata has surface cells."""
return (polydata.GetPolys().GetNumberOfCells() > 0 or
polydata.GetStrips().GetNumberOfCells() > 0)
def _subsample_mesh_points(self, polydata: vtk.vtkPolyData, points_array: np.ndarray) -> np.ndarray:
"""Subsample points from mesh surface."""
try:
sample_points = vtk.vtkPolyDataPointSampler()
sample_points.SetInputData(polydata)
sample_points.SetDistance(self.config.subsampling_distance)
sample_points.Update()
sampled_polydata = sample_points.GetOutput()
if sampled_polydata.GetNumberOfPoints() > 0:
return vtk_to_numpy(sampled_polydata.GetPoints().GetData())
except Exception as e:
print(f"Warning: Point sampling failed: {e}")
return points_array
def _extract_basic_features(self, points_array: np.ndarray, num_points: int, polydata: vtk.vtkPolyData) -> Dict[str, float]:
"""Extract basic geometric features."""
features = {}
# Bounding box
bounds = polydata.GetBounds()
x_length = bounds[1] - bounds[0]
y_length = bounds[3] - bounds[2]
z_length = bounds[5] - bounds[4]
features['x_length'] = x_length
features['y_length'] = y_length
features['z_length'] = z_length
# Aspect ratios
features['aspect_ratio_xy'] = x_length / y_length if y_length > 0 else 0
features['aspect_ratio_xz'] = x_length / z_length if z_length > 0 else 0
features['aspect_ratio_yz'] = y_length / z_length if z_length > 0 else 0
return features
def _extract_shape_features(self, points_array: np.ndarray) -> Dict[str, float]:
"""Extract shape features using PCA."""
features = {}
try:
# Center the points
centered_points = points_array - np.mean(points_array, axis=0)
# Compute covariance matrix
cov_matrix = np.cov(centered_points.T)
# Eigenvalues and eigenvectors
eigenvalues, eigenvectors = np.linalg.eig(cov_matrix)
eigenvalues = np.sort(eigenvalues)[::-1]
if len(eigenvalues) >= 3 and eigenvalues[0] > 0:
# Shape descriptors
features['linearity'] = (eigenvalues[0] - eigenvalues[1]) / eigenvalues[0]
features['planarity'] = (eigenvalues[1] - eigenvalues[2]) / eigenvalues[0]
features['sphericity'] = eigenvalues[2] / eigenvalues[0]
features['anisotropy'] = (eigenvalues[0] - eigenvalues[2]) / eigenvalues[0]
features['omnivariance'] = (eigenvalues[0] * eigenvalues[1] * eigenvalues[2])**(1/3)
# Eigenentropy
normalized_eigenvals = eigenvalues / np.sum(eigenvalues)
features['eigenentropy'] = -np.sum(normalized_eigenvals * np.log(normalized_eigenvals + 1e-10))
# Change of curvature
features['change_of_curvature'] = eigenvalues[2] / (eigenvalues[0] + eigenvalues[1] + eigenvalues[2])
# Add PCA eigenvalues (PhysicsNeMo style)
features['pca_eigenvalue_1'] = eigenvalues[0]
features['pca_eigenvalue_2'] = eigenvalues[1]
features['pca_eigenvalue_3'] = eigenvalues[2]
else:
# Default values
features.update({
'linearity': 0, 'planarity': 0, 'sphericity': 0,
'anisotropy': 0, 'omnivariance': 0, 'eigenentropy': 0,
'change_of_curvature': 0,
'pca_eigenvalue_1': 0, 'pca_eigenvalue_2': 0, 'pca_eigenvalue_3': 0
})
except Exception as e:
features.update({
'linearity': 0, 'planarity': 0, 'sphericity': 0,
'anisotropy': 0, 'omnivariance': 0, 'eigenentropy': 0,
'change_of_curvature': 0,
'pca_eigenvalue_1': 0, 'pca_eigenvalue_2': 0, 'pca_eigenvalue_3': 0
})
return features
class DataLoader:
"""Data loading and preprocessing following PhysicsNeMo methodology."""
def __init__(self, config: AnalysisConfig, processor: Optional[PointCloudProcessor] = None):
self.config = config
self.processor = processor if processor is not None else PointCloudProcessor(config)
def load_vtp_file(self, file_path: Path) -> Optional[vtk.vtkPolyData]:
"""Load a VTP file and return VTK polydata object."""
try:
reader = vtk.vtkXMLPolyDataReader()
reader.SetFileName(str(file_path))
reader.Update()
return reader.GetOutput()
except Exception as e:
print(f"Error loading {file_path}: {e}")
return None
def process_single_file(self, file_path: Path) -> Tuple[str, Dict[str, float]]:
"""Process a single VTP file and extract features."""
car_name = file_path.stem
# Load the VTP file
polydata = self.load_vtp_file(file_path)
if polydata is None:
return car_name, {}
# Extract features
features = self.processor.process(polydata)
return car_name, features
def load_dataset(self) -> Tuple[List[str], np.ndarray, List[str]]:
"""Load all VTP files from multiple data folders and extract features."""
all_results = []
dataset_labels = []
for data_folder in self.config.data_folders:
print(f"\nProcessing folder: {data_folder.label} ({data_folder.path})")
# Find all VTP files in this folder
folder_path = Path(data_folder.path)
if not folder_path.exists():
print(f"Warning: Folder {data_folder.path} does not exist, skipping...")
continue
vtp_files = list(folder_path.glob("**/*.vtp"))
print(f"Found {len(vtp_files)} VTP files ")
# load train names
if data_folder.train_path:
with open(data_folder.train_path, 'r') as f:
train_names = [line.strip() for line in f if line.strip()]
train_set = set(train_names)
# Keep only the vtp files that match train_names
vtp_files = [vtp for vtp in vtp_files if vtp.stem in train_set]
print(f"Kept {len(vtp_files)} VTP files in {data_folder.label}")
# Limit number of files if specified
if self.config.max_files_per_folder > 0:
vtp_files = vtp_files[:self.config.max_files_per_folder]
if len(vtp_files) == 0:
print(f"No VTP files found in {data_folder.label}, skipping...")
continue
# Process files with progress bar
if self.config.use_multiprocessing:
print(f"Processing files using {self.config.num_workers} workers...")
with mp.Pool(self.config.num_workers) as pool:
results = list(tqdm(
pool.imap(self.process_single_file, vtp_files),
total=len(vtp_files),
desc=f"Loading {data_folder.label}",
unit="files"
))
else:
print("Processing files sequentially...")
results = []
for file_path in tqdm(vtp_files, desc=f"Loading {data_folder.label}", unit="files"):
result = self.process_single_file(file_path)
results.append(result)
# Add dataset labels to results
for car_name, features in results:
if features: # Only include files that were successfully processed
all_results.append((car_name, features))
# Use the base folder label for all files, regardless of subdirectory
dataset_labels.append(data_folder.label)
print(f"Successfully processed {len([r for r in results if r[1]])} files from {data_folder.label}")
if len(all_results) == 0:
print("No valid files processed from any folder!")
return [], np.array([]), []
# Extract car names and features
car_names = [name for name, _ in all_results]
all_features = [features for _, features in all_results]
# Convert to DataFrame for easier handling
df = pd.DataFrame(all_features)
# Fill NaN values with 0
df = df.fillna(0)
# Convert to numpy array
features_array = df.values
print(f"\nTotal successfully processed: {len(car_names)} files")
print(f"Extracted {features_array.shape[1]} features per file")
print(f"Dataset distribution: {dict(zip(*np.unique(dataset_labels, return_counts=True)))}")
return car_names, features_array, dataset_labels
def load_test_dataset(self) -> Tuple[List[str], np.ndarray, List[str]]:
"""Load test dataset from test_path configuration."""
all_results = []
dataset_labels = []
for data_folder in self.config.data_folders:
if not data_folder.test_path:
continue
print(f"\nProcessing test folder: {data_folder.label} ({data_folder.test_path})")
# Load test names
with open(data_folder.test_path, 'r') as f:
test_names = [line.strip() for line in f if line.strip()]
test_set = set(test_names)
# Find all VTP files in the test folder
folder_path = Path(data_folder.path)
if not folder_path.exists():
print(f"Warning: Test folder {data_folder.path} does not exist, skipping...")
continue
vtp_files = list(folder_path.glob("**/*.vtp"))
print(f"Found {len(vtp_files)} VTP files in test folder")
# Keep only the vtp files that match test_names
vtp_files = [vtp for vtp in vtp_files if vtp.stem in test_set]
print(f"Kept {len(vtp_files)} VTP files for testing")
if len(vtp_files) == 0:
print(f"No matching test VTP files found, skipping...")
continue
# Process files with progress bar
if self.config.use_multiprocessing:
print(f"Processing test files using {self.config.num_workers} workers...")
with mp.Pool(self.config.num_workers) as pool:
results = list(tqdm(
pool.imap(self.process_single_file, vtp_files),
total=len(vtp_files),
desc=f"Loading test {data_folder.label}",
unit="files"
))
else:
print("Processing test files sequentially...")
results = []
for file_path in tqdm(vtp_files, desc=f"Loading test {data_folder.label}", unit="files"):
result = self.process_single_file(file_path)
results.append(result)
# Add dataset labels to results
for car_name, features in results:
if features: # Only include files that were successfully processed
all_results.append((car_name, features))
dataset_labels.append(f"{data_folder.label}_test")
print(f"Successfully processed {len([r for r in results if r[1]])} test files from {data_folder.label}")
if len(all_results) == 0:
print("No valid test files processed from any folder!")
return [], np.array([]), []
# Extract car names and features
car_names = [name for name, _ in all_results]
all_features = [features for _, features in all_results]
# Convert to DataFrame for easier handling
df = pd.DataFrame(all_features)
# Fill NaN values with 0
df = df.fillna(0)
# Convert to numpy array
features_array = df.values
print(f"\nTotal successfully processed test files: {len(car_names)}")
print(f"Extracted {features_array.shape[1]} features per test file")
return car_names, features_array, dataset_labels
class DimensionalityReducer:
"""Dimensionality reduction following PhysicsNeMo methodology."""
def __init__(self, config: AnalysisConfig):
self.config = config
# Choose scaler based on config
if config.scaling_method == "robust":
self.scaler = RobustScaler() # More robust to outliers than StandardScaler
elif config.scaling_method == "standard":
self.scaler = StandardScaler()
else:
self.scaler = RobustScaler() # Default to robust
def create_umap_embedding(self, features: np.ndarray) -> np.ndarray:
"""Create UMAP embedding from features."""
print("Standardizing features...")
# Standardize features
features_scaled = self.scaler.fit_transform(features)
print("Creating UMAP embedding...")
# Create UMAP embedding
reducer = umap.UMAP(
n_components=self.config.n_components,
n_neighbors=self.config.n_neighbors,
min_dist=self.config.min_dist,
metric=self.config.metric,
random_state=self.config.random_state,
n_epochs=self.config.n_epochs,
learning_rate=self.config.learning_rate,
verbose=True
)
# Perform the embedding
embedding = reducer.fit_transform(features_scaled)
# Store the reducer for later use
self.reducer = reducer
return embedding
class ConfidenceScorer:
"""Confidence scoring for new geometries using HDBSCAN density-based probabilities."""
def __init__(self, config: AnalysisConfig):
self.config = config
self.training_embedding = None
self.training_names = None
self.umap_reducer = None
self.scaler = None
self.hdbscan_clusterer = None
def fit_training_data(self, training_features: np.ndarray, training_names: List[str],
umap_reducer, scaler, hdbscan_clusterer=None) -> None:
"""Fit the confidence scorer on training data."""
self.training_embedding = umap_reducer.transform(scaler.transform(training_features))
self.training_names = training_names
self.umap_reducer = umap_reducer
self.scaler = scaler
self.hdbscan_clusterer = hdbscan_clusterer
print(f"β
Fitted confidence scorer on {len(training_names)} training samples")
def compute_confidence_scores(self, test_features: np.ndarray, test_names: List[str],
k_neighbors: int = 5) -> Tuple[np.ndarray, np.ndarray]:
"""Compute confidence scores for test geometries using HDBSCAN density-based probabilities."""
if self.training_embedding is None:
raise ValueError("Confidence scorer not fitted. Call fit_training_data() first.")
print(f"π Computing HDBSCAN density-based confidence scores for {len(test_names)} test geometries...")
# Transform test features to UMAP space
test_features_scaled = self.scaler.transform(test_features)
test_embedding = self.umap_reducer.transform(test_features_scaled)
# Use HDBSCAN density-based confidence if clusterer is available
if self.hdbscan_clusterer is not None and HDBSCAN_AVAILABLE:
# Use HDBSCAN approximate prediction for density-based confidence
pred_labels, pred_probs = prediction.approximate_predict(self.hdbscan_clusterer, test_embedding)
confidence_scores = pred_probs
print(f"β
Computed HDBSCAN density-based confidence scores:")
print(f" - Mean confidence: {np.mean(confidence_scores):.4f}")
return test_embedding, confidence_scores
def create_confidence_visualization(self, test_embedding: np.ndarray, test_names: List[str],
confidence_scores: np.ndarray,
save_path: str = "files/confidence_visualization.png") -> None:
"""Create visualization showing training vs test points with confidence scores."""
plt.figure(figsize=(12, 10))
# Plot training points in light gray
plt.scatter(self.training_embedding[:, 0], self.training_embedding[:, 1],
c='lightgray', s=30, alpha=0.6, label='Training Data', edgecolors='black', linewidth=0.5)
# Plot test points colored by confidence score
plt.scatter(test_embedding[:, 0], test_embedding[:, 1],
c=confidence_scores, s=100, alpha=0.8,
cmap='RdYlGn', vmin=0, vmax=1,
label='Test Data', edgecolors='black', linewidth=1)
# Add colorbar
# cbar = plt.colorbar(scatter)
# cbar.set_label('Confidence Score', fontsize=12)
# Add labels for test points
for i, name in enumerate(test_names):
plt.annotate(name, (test_embedding[i, 0], test_embedding[i, 1]),
xytext=(5, 5), textcoords='offset points',
fontsize=8, alpha=0.8, fontweight='bold')
plt.xlabel('UMAP Component 1', fontsize=12)
plt.ylabel('UMAP Component 2', fontsize=12)
plt.title('Confidence Scoring: Training vs Test Geometries', fontsize=16, fontweight='bold')
plt.legend()
plt.grid(True, alpha=0.3)
# Save plot
os.makedirs(os.path.dirname(save_path), exist_ok=True)
plt.savefig(save_path, dpi=300, bbox_inches='tight')
print(f"Confidence visualization saved to: {save_path}")
plt.show()
class Clusterer:
"""HDBSCAN clustering for UMAP embeddings following PhysicsNeMo methodology."""
def __init__(self, config: AnalysisConfig):
self.config = config
def cluster_embedding(self, embedding: np.ndarray) -> Tuple[np.ndarray, np.ndarray, Any]:
"""Perform HDBSCAN clustering on UMAP embedding."""
print("π Performing HDBSCAN clustering on UMAP embedding...")
# Calculate adaptive min_cluster_size based on number of points
num_points = len(embedding)
# adaptive_min_cluster_size = min(self.config.min_cluster_size, max(2, num_points // 10))
# For very small datasets, use even smaller cluster size
# if num_points < 20:
# adaptive_min_cluster_size = 2
# elif num_points < 50:
# adaptive_min_cluster_size = max(2, num_points // 5)
print(f"π Clustering parameters:")
print(f" - Number of points: {num_points}")
print(f" - Config min_cluster_size: {self.config.min_cluster_size}")
# print(f" - Adaptive min_cluster_size: {adaptive_min_cluster_size}")
try:
if HDBSCAN_AVAILABLE:
clusterer = hdbscan.HDBSCAN(
min_cluster_size=self.config.min_cluster_size,
metric=self.config.metric,
cluster_selection_epsilon=self.config.cluster_selection_epsilon,
allow_single_cluster=self.config.allow_single_cluster,
prediction_data=True # Enable prediction data for approximate_predict
)
cluster_labels = clusterer.fit_predict(embedding)
print("β
Used HDBSCAN (CPU)")
else:
raise Exception("HDBSCAN not available")
# Identify outliers (label = -1)
outlier_mask = cluster_labels == -1
outliers = np.where(outlier_mask)[0]
print(f"π Clustering Results:")
print(f" - Total points: {len(cluster_labels)}")
print(f" - Number of clusters: {len(set(cluster_labels)) - (1 if -1 in cluster_labels else 0)}")
print(f" - Outliers: {len(outliers)} ({len(outliers)/len(cluster_labels)*100:.1f}%)")
# Print density statistics
if hasattr(clusterer, 'probabilities_') and clusterer.probabilities_ is not None:
print(f" - Mean cluster probability: {np.mean(clusterer.probabilities_):.4f}")
print(f" - Min cluster probability: {np.min(clusterer.probabilities_):.4f}")
print(f" - Max cluster probability: {np.max(clusterer.probabilities_):.4f}")
# Check if prediction data is available
if hasattr(clusterer, 'prediction_data_') and clusterer.prediction_data_ is not None:
print(f" - Prediction data: β
Available for approximate_predict")
else:
print(f" - Prediction data: β Not available - will use distance-based fallback")
# Check if any clusters were found
unique_labels = set(cluster_labels)
num_clusters = len(unique_labels) - (1 if -1 in unique_labels else 0)
if num_clusters == 0:
print("β οΈ No clusters found! Attempting fallback with more permissive parameters...")
return self._fallback_clustering(embedding)
return cluster_labels, outliers, clusterer
except Exception as e:
print(f"β Clustering failed: {e}")
# def _fallback_clustering(self, embedding: np.ndarray) -> Tuple[np.ndarray, np.ndarray, Any]:
# """Fallback clustering with more permissive parameters."""
# print("π Attempting fallback with more permissive parameters...")
# # Try multiple fallback strategies
# fallback_strategies = [
# # Strategy 1: Very permissive HDBSCAN
# {
# 'min_cluster_size': 2,
# 'cluster_selection_epsilon': 0.0,
# 'allow_single_cluster': True,
# 'min_samples': 1
# },
# # Strategy 2: Even more permissive
# {
# 'min_cluster_size': 2,
# 'cluster_selection_epsilon': 0.0,
# 'allow_single_cluster': True,
# 'min_samples': 1,
# 'cluster_selection_method': 'eom'
# },
# # Strategy 3: Single cluster fallback
# {
# 'min_cluster_size': 2,
# 'cluster_selection_epsilon': 0.0,
# 'allow_single_cluster': True,
# 'min_samples': 1,
# 'cluster_selection_method': 'leaf'
# }
# ]
# for i, strategy in enumerate(fallback_strategies):
# try:
# if HDBSCAN_AVAILABLE:
# print(f" Trying fallback strategy {i+1}...")
# fallback_clusterer = hdbscan.HDBSCAN(
# prediction_data=True,
# **strategy
# )
# fallback_labels = fallback_clusterer.fit_predict(embedding)
# # Check if clusters were found
# unique_labels = set(fallback_labels)
# num_clusters = len(unique_labels) - (1 if -1 in unique_labels else 0)
# if num_clusters > 0:
# print(f"β
Fallback strategy {i+1} succeeded with {num_clusters} clusters")
# outliers = np.where(fallback_labels == -1)[0]
# return fallback_labels, outliers, fallback_clusterer
# else:
# print(f" Strategy {i+1} found no clusters, trying next...")
# except Exception as e:
# print(f" Strategy {i+1} failed: {e}")
# continue
# # If all strategies fail, create a single cluster with all points
# print("β οΈ All fallback strategies failed, creating single cluster with all points...")
# try:
# # Create a dummy clusterer that treats all points as one cluster
# dummy_clusterer = hdbscan.HDBSCAN(
# min_cluster_size=1,
# cluster_selection_epsilon=0.0,
# allow_single_cluster=True,
# prediction_data=True
# )
# # Force all points into cluster 0
# dummy_labels = np.zeros(len(embedding), dtype=int)
# print("β
Created single cluster fallback")
# return dummy_labels, np.array([]), dummy_clusterer
# except Exception as final_e:
# print(f"β Final fallback also failed: {final_e}")
# # Return no clusters, no outliers, and None clusterer
# return np.zeros(len(embedding), dtype=int), np.array([]), None
class Visualizer:
"""Visualization methodology."""
def __init__(self, config: AnalysisConfig):
self.config = config
def create_umap_visualization(self, embedding: np.ndarray, car_names: List[str],
dataset_labels: List[str] = None, cluster_labels: np.ndarray = None,
outliers: np.ndarray = None) -> None:
"""Create and save UMAP visualization with clustering."""
plt.figure(figsize=self.config.figsize)
if cluster_labels is not None and len(cluster_labels) > 0:
# Create visualization with clustering
unique_clusters = sorted([c for c in set(cluster_labels) if c != -1])
unique_datasets = list(set(dataset_labels)) if dataset_labels else ['Dataset']
# Color maps
cluster_colors = plt.cm.tab10(np.linspace(0, 1, len(unique_clusters)))
dataset_colors = plt.cm.Set1(np.linspace(0, 1, len(unique_datasets)))
# Plot clusters
for cluster_idx, cluster_id in enumerate(unique_clusters):
cluster_mask = cluster_labels == cluster_id
if dataset_labels is not None:
# Plot each dataset separately within cluster
for dataset_idx, dataset in enumerate(unique_datasets):
dataset_mask = np.array(dataset_labels) == dataset
combined_mask = cluster_mask & dataset_mask
if np.any(combined_mask):
plt.scatter(
embedding[combined_mask, 0],
embedding[combined_mask, 1],
s=self.config.point_size,
alpha=self.config.alpha,
c=[cluster_colors[cluster_idx]],
marker='o',
edgecolors='black',
linewidth=0.5,
label=f'{dataset} - Cluster {cluster_id}' if cluster_idx == 0 else ""
)
else:
# Plot cluster without dataset distinction
plt.scatter(
embedding[cluster_mask, 0],
embedding[cluster_mask, 1],
s=self.config.point_size,
alpha=self.config.alpha,
c=[cluster_colors[cluster_idx]],
label=f'Cluster {cluster_id}',
edgecolors='black',
linewidth=0.5
)
# Plot outliers
if outliers is not None and len(outliers) > 0:
plt.scatter(
embedding[outliers, 0],
embedding[outliers, 1],
s=self.config.point_size * 1.5,
alpha=self.config.alpha,
c='red',
marker='x',
linewidth=2,
label='Outliers'
)
plt.legend(title='Clusters', bbox_to_anchor=(1.05, 1), loc='upper left')
plt.title('UMAP Visualization with HDBSCAN Clustering', fontsize=16, fontweight='bold')
elif dataset_labels is not None:
# Create visualization with dataset-based coloring (no clustering)
unique_datasets = list(set(dataset_labels))
colors = plt.cm.Set1(np.linspace(0, 1, len(unique_datasets)))
dataset_color_map = dict(zip(unique_datasets, colors))
# Plot each dataset separately
for dataset in unique_datasets:
mask = np.array(dataset_labels) == dataset
plt.scatter(
embedding[mask, 0],
embedding[mask, 1],
s=self.config.point_size,
alpha=self.config.alpha,
c=[dataset_color_map[dataset]],
label=dataset,
edgecolors='black',
linewidth=0.5
)
plt.legend(title='Dataset', bbox_to_anchor=(1.05, 1), loc='upper left')
plt.title('UMAP Visualization - Multi-Dataset Analysis', fontsize=16, fontweight='bold')
else:
# Create scatter plot with index-based coloring
scatter = plt.scatter(
embedding[:, 0],
embedding[:, 1],
s=self.config.point_size,
alpha=self.config.alpha,
c=range(len(car_names)),
cmap=self.config.colormap
)
# Add colorbar
cbar = plt.colorbar(scatter)
cbar.set_label('Model Index', fontsize=12)
plt.title('UMAP Visualization of Dataset', fontsize=16, fontweight='bold')
plt.xlabel('UMAP Component 1', fontsize=12)
plt.ylabel('UMAP Component 2', fontsize=12)
# Add labels if requested
if self.config.show_labels:
for i, name in enumerate(car_names):
plt.annotate(name, (embedding[i, 0], embedding[i, 1]),
xytext=(5, 5), textcoords='offset points',
fontsize=8, alpha=0.7)
plt.tight_layout()
# Save plot if requested
if self.config.save_plot:
os.makedirs(os.path.dirname(self.config.output_path), exist_ok=True)
plt.savefig(self.config.output_path, dpi=self.config.dpi, bbox_inches='tight')
print(f"Plot saved to: {self.config.output_path}")
plt.show()
def create_comparison_plot(self, umap_embedding: np.ndarray, pca_embedding: np.ndarray, car_names: List[str], dataset_labels: List[str] = None) -> None:
"""Create comparison plot between UMAP and PCA."""
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 8))
if dataset_labels is not None:
# Create plots with dataset-based coloring
unique_datasets = list(set(dataset_labels))
colors = plt.cm.Set1(np.linspace(0, 1, len(unique_datasets)))
dataset_color_map = dict(zip(unique_datasets, colors))
# UMAP plot
for dataset in unique_datasets:
mask = np.array(dataset_labels) == dataset
ax1.scatter(
umap_embedding[mask, 0],
umap_embedding[mask, 1],
s=self.config.point_size,
alpha=self.config.alpha,
c=[dataset_color_map[dataset]],
label=dataset,
edgecolors='black',
linewidth=0.5
)
ax1.set_title('UMAP Embedding', fontsize=14, fontweight='bold')
ax1.set_xlabel('UMAP Component 1', fontsize=12)
ax1.set_ylabel('UMAP Component 2', fontsize=12)
ax1.legend(title='Dataset', loc='upper right')
# PCA plot
for dataset in unique_datasets:
mask = np.array(dataset_labels) == dataset
ax2.scatter(
pca_embedding[mask, 0],
pca_embedding[mask, 1],
s=self.config.point_size,
alpha=self.config.alpha,
c=[dataset_color_map[dataset]],
label=dataset,
edgecolors='black',
linewidth=0.5
)
ax2.set_title('PCA Embedding', fontsize=14, fontweight='bold')
ax2.set_xlabel('PCA Component 1', fontsize=12)
ax2.set_ylabel('PCA Component 2', fontsize=12)
ax2.legend(title='Dataset', loc='upper right')
else:
# UMAP plot
scatter1 = ax1.scatter(
umap_embedding[:, 0],
umap_embedding[:, 1],
s=self.config.point_size,
alpha=self.config.alpha,
c=range(len(car_names)),
cmap=self.config.colormap
)
ax1.set_title('UMAP Embedding', fontsize=14, fontweight='bold')
ax1.set_xlabel('UMAP Component 1', fontsize=12)
ax1.set_ylabel('UMAP Component 2', fontsize=12)
plt.colorbar(scatter1, ax=ax1, label='Model Index')
# PCA plot
scatter2 = ax2.scatter(
pca_embedding[:, 0],
pca_embedding[:, 1],
s=self.config.point_size,
alpha=self.config.alpha,
c=range(len(car_names)),
cmap=self.config.colormap
)
ax2.set_title('PCA Embedding', fontsize=14, fontweight='bold')
ax2.set_xlabel('PCA Component 1', fontsize=12)
ax2.set_ylabel('PCA Component 2', fontsize=12)
plt.colorbar(scatter2, ax=ax2, label='Model Index')
plt.tight_layout()
plt.show()
def create_feature_analysis(self, features: np.ndarray, car_names: List[str]) -> None:
"""Create feature analysis plots."""
# Create comprehensive feature names list
feature_names = [
'x_length', 'y_length', 'z_length',
'aspect_ratio_xy', 'aspect_ratio_xz', 'aspect_ratio_yz',
'linearity', 'planarity', 'sphericity', 'anisotropy',
'omnivariance', 'eigenentropy', 'change_of_curvature',
'pca_eigenvalue_1', 'pca_eigenvalue_2', 'pca_eigenvalue_3'
]
# Ensure we have the right number of feature names
if len(feature_names) < features.shape[1]:
# Add generic names for any missing features
for i in range(len(feature_names), features.shape[1]):
feature_names.append(f'feature_{i}')
elif len(feature_names) > features.shape[1]:
# Truncate if we have too many names
feature_names = feature_names[:features.shape[1]]
df_features = pd.DataFrame(features, columns=feature_names)
# Correlation heatmap
plt.figure(figsize=(12, 10))
correlation_matrix = df_features.corr()
sns.heatmap(correlation_matrix, annot=True, cmap='coolwarm', center=0, fmt='.2f')
plt.title('Feature Correlation Matrix', fontsize=16, fontweight='bold')
plt.tight_layout()
plt.show()
# Feature distribution
fig, axes = plt.subplots(4, 4, figsize=(16, 16))
axes = axes.ravel()
for i, feature in enumerate(feature_names[:16]):
if i < len(axes):
axes[i].hist(features[:, i], bins=30, alpha=0.7, edgecolor='black')
axes[i].set_title(f'{feature}', fontsize=10)
axes[i].set_xlabel('Value')
axes[i].set_ylabel('Frequency')
plt.suptitle('Feature Distributions', fontsize=16, fontweight='bold')
plt.tight_layout()
plt.show()
class PhysicsNeMoAnalyzer:
"""Main analyzer class following PhysicsNeMo methodology."""
def __init__(self, config_path: str):
"""Initialize with configuration file."""
# Load configuration from YAML file
self.config = AnalysisConfig.from_yaml(config_path)
# Initialize components
self.processor = PointCloudProcessor(self.config)
self.data_loader = DataLoader(self.config, self.processor)
self.dimensionality_reducer = DimensionalityReducer(self.config)
self.clusterer = Clusterer(self.config)
self.visualizer = Visualizer(self.config)
self.confidence_scorer = ConfidenceScorer(self.config)
def run_unified_analysis(self) -> None:
"""Run the unified analysis pipeline: training + confidence scoring in one go."""
print("=" * 60)
print("Unified PhysicsNeMo Analysis Pipeline")
print("=" * 60)
# Step 1: Load training dataset
print("\nπ Step 1/6: Loading training dataset...")
train_names, train_features, train_labels = self.data_loader.load_dataset()
if len(train_names) == 0:
print("β No training data found!")
return
print(f"β
Successfully loaded {len(train_names)} training samples")
# Step 2: Create UMAP embedding on training data
print(f"\nπ§ Step 2/6: Creating UMAP embedding on training data...")
train_embedding = self.dimensionality_reducer.create_umap_embedding(train_features)
print("β
UMAP embedding completed")
# Step 3: Perform HDBSCAN clustering on training embedding
print(f"\nπ Step 3/6: Performing HDBSCAN clustering...")
cluster_labels, outliers, clusterer = self.clusterer.cluster_embedding(train_embedding)
print("β
HDBSCAN clustering completed")
# Step 4: Load test dataset and compute confidence scores
print(f"\nπ Step 4/6: Loading test dataset and computing confidence scores...")
test_names, test_features, test_labels = self.data_loader.load_test_dataset()
if len(test_names) > 0:
print(f"β
Successfully loaded {len(test_names)} test samples")
# Fit confidence scorer and compute scores
self.confidence_scorer.fit_training_data(
train_features, train_names,
self.dimensionality_reducer.reducer,
self.dimensionality_reducer.scaler,
clusterer
)
test_embedding, confidence_scores = self.confidence_scorer.compute_confidence_scores(
test_features, test_names, k_neighbors=4
)
print("β
Confidence scores computed")
else:
print("β οΈ No test data found, skipping confidence scoring")
test_embedding, confidence_scores, mean_confidence_scores = None, None, None
# Step 5: Create visualizations
print(f"\nπ¨ Step 5/6: Creating visualizations...")
self.visualizer.create_umap_visualization(train_embedding, train_names, train_labels, cluster_labels, outliers)
self.visualizer.create_feature_analysis(train_features, train_names)
if test_embedding is not None:
self.confidence_scorer.create_confidence_visualization(
test_embedding, test_names, confidence_scores
)
print("β
Visualizations completed")
# Step 6: Save all results
print(f"\nπΎ Step 6/6: Saving all results...")
self._save_results(train_embedding, train_features, train_names, train_labels, cluster_labels, outliers, clusterer)
if test_embedding is not None:
self._save_confidence_results(
test_embedding, test_features, test_names, test_labels,
confidence_scores
)
print("β
All results saved")
print("\n" + "=" * 60)
print("π Unified analysis complete!")
print("=" * 60)
def run_analysis(self) -> None:
"""Run the complete analysis pipeline (legacy method)."""
print("=" * 60)
print("PhysicsNeMo-Style Car Dataset Analysis")
print("=" * 60)
# Step 1: Load dataset
print("\nπ Step 1/5: Loading dataset...")
car_names, features, dataset_labels = self.data_loader.load_dataset()
if len(car_names) == 0:
print("β No valid files found!")
return
print(f"β
Successfully loaded {len(car_names)} models")
# Step 2: Create UMAP embedding
print(f"\nπ§ Step 2/5: Creating UMAP embedding...")
umap_embedding = self.dimensionality_reducer.create_umap_embedding(features)
print("β
UMAP embedding completed")
# Step 3: Perform HDBSCAN clustering on UMAP embedding
print(f"\nπ Step 3/5: Performing HDBSCAN clustering...")
cluster_labels, outliers, clusterer = self.clusterer.cluster_embedding(umap_embedding)
print("β
HDBSCAN clustering completed")
# Step 4: Create visualizations
print(f"\nπ¨ Step 4/5: Creating visualizations...")
self.visualizer.create_umap_visualization(umap_embedding, car_names, dataset_labels, cluster_labels, outliers)
self.visualizer.create_feature_analysis(features, car_names)
print("β
Visualizations completed")
# Step 5: Save results
print(f"\nπΎ Step 5/5: Saving results...")
self._save_results(umap_embedding, features, car_names, dataset_labels, cluster_labels, outliers, clusterer)
print("β
Results saved")
print("\n" + "=" * 60)
print("π Analysis complete!")
print("=" * 60)
def run_confidence_scoring(self) -> None:
"""Run the confidence scoring pipeline for test data."""
print("=" * 60)
print("Confidence Scoring Pipeline")
print("=" * 60)
# Step 1: Load training data
print("\nπ Step 1/5: Loading training dataset...")
train_names, train_features, train_labels = self.data_loader.load_dataset()
if len(train_names) == 0:
print("β No training data found!")
return
print(f"β
Successfully loaded {len(train_names)} training samples")
# Step 2: Fit UMAP on training data only
print(f"\nπ§ Step 2/5: Fitting UMAP on training data...")
umap_embedding = self.dimensionality_reducer.create_umap_embedding(train_features)
print("β
UMAP fitted on training data")
# Step 2.5: Perform HDBSCAN clustering on training embedding
print(f"\nπ Step 2.5/5: Performing HDBSCAN clustering...")
cluster_labels, outliers, clusterer = self.clusterer.cluster_embedding(umap_embedding)
print("β
HDBSCAN clustering completed")
# Step 3: Load test data
print(f"\nπ Step 3/5: Loading test dataset...")
test_names, test_features, test_labels = self.data_loader.load_test_dataset()
if len(test_names) == 0:
print("β No test data found!")
return
print(f"β
Successfully loaded {len(test_names)} test samples")
# Step 4: Fit confidence scorer and compute scores
print(f"\nπ Step 4/5: Computing confidence scores...")
self.confidence_scorer.fit_training_data(
train_features, train_names,
self.dimensionality_reducer.reducer,
self.dimensionality_reducer.scaler,
clusterer
)
test_embedding, confidence_scores = self.confidence_scorer.compute_confidence_scores(
test_features, test_names, k_neighbors=4
)
print("β
Confidence scores computed")
# Step 5: Create visualizations and save results
print(f"\nπ¨ Step 5/5: Creating visualizations and saving results...")
self.confidence_scorer.create_confidence_visualization(
test_embedding, test_names, confidence_scores
)
# Save confidence results
self._save_confidence_results(
test_embedding, test_features, test_names, test_labels,
confidence_scores
)
print("β
Results saved")
print("\n" + "=" * 60)
print("π Confidence scoring complete!")
print("=" * 60)
def _save_confidence_results(self, test_embedding: np.ndarray, test_features: np.ndarray,
test_names: List[str], test_labels: List[str],
confidence_scores: np.ndarray) -> None:
"""Save confidence scoring results to files."""
# Generate dynamic file names
unique_datasets = list(set(test_labels))
if len(unique_datasets) == 1:
base_name = unique_datasets[0].replace(" ", "_").lower()
else:
base_name = "combined_test"
# Save test embedding
# test_embedding_path = f"files/{base_name}_test_embedding.npy"
# os.makedirs(os.path.dirname(test_embedding_path), exist_ok=True)
# np.save(test_embedding_path, test_embedding)
# print(f"Test embedding saved to: {test_embedding_path}")
# # Save test features
# test_features_path = f"files/{base_name}_test_features.npy"
# np.save(test_features_path, test_features)
# print(f"Test features saved to: {test_features_path}")
# Save test names
# test_names_path = f"files/{base_name}_test_names.npy"
# np.save(test_names_path, test_names)
# print(f"Test names saved to: {test_names_path}")
# Save confidence scores
# confidence_path = f"files/{base_name}_confidence_scores.npy"
# np.save(confidence_path, confidence_scores)
# print(f"Confidence scores saved to: {confidence_path}")
# Save mean confidence scores
# mean_confidence_path = f"files/{base_name}_mean_confidence_scores.npy"
# np.save(mean_confidence_path, mean_confidence_scores)
# print(f"Mean confidence scores saved to: {mean_confidence_path}")
# Create a summary report
summary_path = f"files/{base_name}_confidence_summary.txt"
with open(summary_path, 'w') as f:
f.write("Confidence Scoring Summary\n")
f.write("=" * 30 + "\n\n")
f.write(f"Number of test samples: {len(test_names)}\n")
f.write(f"Confidence score range: {np.min(confidence_scores):.4f} - {np.max(confidence_scores):.4f}\n")
f.write(f"Mean confidence score: {np.mean(confidence_scores):.4f}\n")
f.write(f"Std confidence score: {np.std(confidence_scores):.4f}\n\n")
f.write("Individual Results:\n")
f.write("-" * 20 + "\n")
for i, (name, conf) in enumerate(zip(test_names, confidence_scores)):
f.write(f"{i+1:3d}. {name:20s} | Confidence: {conf:.4f}\n")
print(f"Summary report saved to: {summary_path}")
def _save_results(self, embedding: np.ndarray, features: np.ndarray, car_names: List[str],
dataset_labels: List[str], cluster_labels: np.ndarray = None, outliers: np.ndarray = None, clusterer: Any = None) -> None:
"""Save results to files."""
# Generate dynamic file names based on dataset labels
unique_datasets = list(set(dataset_labels))
if len(unique_datasets) == 1:
# Single dataset - use the label name
base_name = unique_datasets[0].replace(" ", "_").lower()
else:
# Multiple datasets - use "combined"
base_name = "combined"
# Generate file paths (save to files/ directory)
embedding_path = f"files/{base_name}_umap_embedding.npy"
features_path = f"files/{base_name}_features.npy"
labels_path = f"files/{base_name}_names.npy"
dataset_labels_path = f"files/{base_name}_dataset_labels.npy"
# Save embedding
if self.config.save_embedding:
# Only create directory if there's a directory path
dir_path = os.path.dirname(embedding_path)
if dir_path:
os.makedirs(dir_path, exist_ok=True)
np.save(embedding_path, embedding)
print(f"Embedding saved to: {embedding_path}")
# Save features
if self.config.save_features:
# Only create directory if there's a directory path
dir_path = os.path.dirname(features_path)
if dir_path:
os.makedirs(dir_path, exist_ok=True)
np.save(features_path, features)
print(f"Features saved to: {features_path}")
# Save labels
if self.config.save_labels:
# Only create directory if there's a directory path
dir_path = os.path.dirname(labels_path)
if dir_path:
os.makedirs(dir_path, exist_ok=True)
np.save(labels_path, car_names)
print(f"Names saved to: {labels_path}")
# Save dataset labels
# Only create directory if there's a directory path
dir_path = os.path.dirname(dataset_labels_path)
if dir_path:
os.makedirs(dir_path, exist_ok=True)
np.save(dataset_labels_path, dataset_labels)
print(f"Dataset labels saved to: {dataset_labels_path}")
# Save UMAP reducer, scaler, and clusterer for reuse in Gradio demo
self._save_umap_components(base_name, clusterer)
# Save clustering results
# if cluster_labels is not None:
# cluster_labels_path = self.config.labels_path.replace('.npy', '_cluster_labels.npy')
# os.makedirs(os.path.dirname(cluster_labels_path), exist_ok=True)
# np.save(cluster_labels_path, cluster_labels)
# print(f"Cluster labels saved to: {cluster_labels_path}")
# if outliers is not None and len(outliers) > 0:
# outliers_path = self.config.labels_path.replace('.npy', '_outliers.npy')
# os.makedirs(os.path.dirname(outliers_path), exist_ok=True)
# np.save(outliers_path, outliers)
# print(f"Outliers saved to: {outliers_path}")
def _save_umap_components(self, base_name: str, clusterer=None) -> None:
"""Save UMAP reducer, scaler, and clusterer for reuse in Gradio demo."""
import pickle
# Save UMAP reducer
reducer_path = f"files/{base_name}_umap_reducer.pkl"
dir_path = os.path.dirname(reducer_path)
if dir_path:
os.makedirs(dir_path, exist_ok=True)
with open(reducer_path, 'wb') as f:
pickle.dump(self.dimensionality_reducer.reducer, f)
print(f"UMAP reducer saved to: {reducer_path}")
# Save scaler
scaler_path = f"files/{base_name}_scaler.pkl"
with open(scaler_path, 'wb') as f:
pickle.dump(self.dimensionality_reducer.scaler, f)
print(f"Scaler saved to: {scaler_path}")
# Save HDBSCAN clusterer if provided
if clusterer is not None:
clusterer_path = f"files/{base_name}_hdbscan_clusterer.pkl"
with open(clusterer_path, 'wb') as f:
pickle.dump(clusterer, f)
print(f"HDBSCAN clusterer saved to: {clusterer_path}")
def main():
"""Main function to run the PhysicsNeMo-style analysis."""
import sys
config_path = "/raid/ansysai/udbhav/alphaLPFM/similarity/umap/config.yaml"
if not os.path.exists(config_path):
print(f"Configuration file not found: {config_path}")
return
# Create analyzer
analyzer = PhysicsNeMoAnalyzer(config_path)
# Check command line arguments for mode
if len(sys.argv) > 1:
if sys.argv[1] == "confidence":
print("Running confidence scoring pipeline only...")
analyzer.run_confidence_scoring()
elif sys.argv[1] == "train":
print("Running training analysis only...")
analyzer.run_analysis()
else:
print(f"Unknown argument: {sys.argv[1]}")
print("Usage: python run_umap.py [train|confidence]")
print(" - No argument: Run unified pipeline (training + confidence scoring)")
print(" - train: Run training analysis only")
print(" - confidence: Run confidence scoring only")
else:
print("Running unified pipeline (training + confidence scoring)...")
analyzer.run_unified_analysis()
if __name__ == "__main__":
main() |