Skip to content

Commit aaf4f5f

Browse files
TasfinMahmudTasfinMahmud
committed
fix: run ruff format on validate_pu.py
1 parent a66c313 commit aaf4f5f

1 file changed

Lines changed: 109 additions & 66 deletions

File tree

‎validate_pu.py‎

Lines changed: 109 additions & 66 deletions
Original file line numberDiff line numberDiff line change
@@ -1,30 +1,32 @@
1-
import os
21
import glob
2+
import os
3+
34
import numpy as np
45
import scipy.io as sio
56

67
from cnsd import Dataset
78
from cnsd.diagnosis.system import CNSD
89
from cnsd.physics import PhysicsConfig
910

10-
def load_pu_domain_split(data_dir=r"E:\301\PU-dataset", window_size=8192):
11+
12+
def load_pu_domain_split(data_dir=r'E:\301\PU-dataset', window_size=8192):
1113
"""
1214
Loads authentic PU dataset and strictly splits by RPM (Domain Shift).
1315
Train: N09 (900 RPM)
1416
Test/Calib: N15 (1500 RPM)
1517
"""
1618
X_train, y_train, cond_train = [], [], []
1719
X_target, y_target, cond_target = [], [], []
18-
19-
categories = [("K0*", 0), ("KA*", 1), ("KI*", 2)]
20-
20+
21+
categories = [('K0*', 0), ('KA*', 1), ('KI*', 2)]
22+
2123
for prefix, label in categories:
22-
pattern = os.path.join(data_dir, prefix, "*.mat")
24+
pattern = os.path.join(data_dir, prefix, '*.mat')
2325
files = glob.glob(pattern)
2426
for fpath in files:
2527
fname = os.path.basename(fpath)
2628
key = fname.replace('.mat', '')
27-
29+
2830
rpm_code = fname.split('_')[0]
2931
if rpm_code == 'N09':
3032
rpm = 900.0
@@ -33,30 +35,30 @@ def load_pu_domain_split(data_dir=r"E:\301\PU-dataset", window_size=8192):
3335
rpm = 1500.0
3436
is_train = False
3537
else:
36-
continue # ignore other speeds if they exist
37-
38+
continue # ignore other speeds if they exist
39+
3840
try:
3941
mat = sio.loadmat(fpath)
4042
if key not in mat:
4143
continue
42-
43-
y_struct = mat[key]['Y'][0,0]
44-
44+
45+
y_struct = mat[key]['Y'][0, 0]
46+
4547
vib_idx = -1
4648
for i in range(y_struct['Name'].shape[1]):
47-
if 'vibration' in str(y_struct['Name'][0,i][0]).lower():
49+
if 'vibration' in str(y_struct['Name'][0, i][0]).lower():
4850
vib_idx = i
4951
break
50-
52+
5153
if vib_idx == -1:
5254
continue
53-
54-
sig = y_struct['Data'][0,vib_idx].flatten()
55-
55+
56+
sig = y_struct['Data'][0, vib_idx].flatten()
57+
5658
for i in range(0, len(sig) - window_size, window_size):
57-
segment = sig[i:i+window_size]
59+
segment = sig[i : i + window_size]
5860
segment = (segment - np.mean(segment)) / (np.std(segment) + 1e-8)
59-
61+
6062
if is_train:
6163
X_train.append(segment)
6264
y_train.append(label)
@@ -65,12 +67,16 @@ def load_pu_domain_split(data_dir=r"E:\301\PU-dataset", window_size=8192):
6567
X_target.append(segment)
6668
y_target.append(label)
6769
cond_target.append(rpm)
68-
70+
6971
except Exception as e:
70-
print(f"Error loading {fname}: {e}")
71-
72-
return (np.array(X_train, dtype=np.float32), np.array(y_train), np.array(cond_train)), \
73-
(np.array(X_target, dtype=np.float32), np.array(y_target), np.array(cond_target))
72+
print(f'Error loading {fname}: {e}')
73+
74+
return (np.array(X_train, dtype=np.float32), np.array(y_train), np.array(cond_train)), (
75+
np.array(X_target, dtype=np.float32),
76+
np.array(y_target),
77+
np.array(cond_target),
78+
)
79+
7480

7581
def headline_accuracy_by_verdict(report, y_true):
7682
pred = np.array([r['predicted_class'] for r in report.records])
@@ -83,95 +89,132 @@ def headline_accuracy_by_verdict(report, y_true):
8389
out[v] = {'n': int(m.sum()), 'cnn_accuracy': float(correct[m].mean())}
8490
return out
8591

86-
if __name__ == "__main__":
87-
print("Loading Authentic PU dataset (Cross-Domain RPM Split)...")
92+
93+
if __name__ == '__main__':
94+
print('Loading Authentic PU dataset (Cross-Domain RPM Split)...')
8895
(X_train, y_train, cond_train), (X_target, y_target, cond_target) = load_pu_domain_split()
89-
96+
9097
# Split target domain into Calib (50%) and Test (50%)
9198
indices = np.arange(len(y_target))
9299
np.random.shuffle(indices)
93100
calib_size = len(indices) // 2
94-
101+
95102
calib_idx = indices[:calib_size]
96103
test_idx = indices[calib_size:]
97-
104+
98105
X_calib, y_calib, cond_calib = X_target[calib_idx], y_target[calib_idx], cond_target[calib_idx]
99106
X_test, y_test, cond_test = X_target[test_idx], y_target[test_idx], cond_target[test_idx]
100-
107+
101108
# Shuffle train set
102109
train_indices = np.arange(len(y_train))
103110
np.random.shuffle(train_indices)
104-
X_train, y_train, cond_train = X_train[train_indices], y_train[train_indices], cond_train[train_indices]
105-
106-
print(f"Data split: Train (900 RPM)={len(y_train)} | Calib (1500 RPM)={len(y_calib)} | Test (1500 RPM)={len(y_test)}")
107-
111+
X_train, y_train, cond_train = (
112+
X_train[train_indices],
113+
y_train[train_indices],
114+
cond_train[train_indices],
115+
)
116+
117+
print(
118+
f'Data split: Train (900 RPM)={len(y_train)} | Calib (1500 RPM)={len(y_calib)} | Test (1500 RPM)={len(y_test)}'
119+
)
120+
108121
unique_rpm = set(cond_train).union(set(cond_calib)).union(set(cond_test))
109122
rpm_map = {float(r): float(r) for r in unique_rpm}
110-
123+
111124
pu_physics = PhysicsConfig(
112125
bearing={'n_balls': 8, 'd_ball': 6.75, 'd_pitch': 28.5, 'contact_angle': 0.0},
113126
cond_to_rpm=rpm_map,
114127
fs=64000,
115-
name='PU-6203'
128+
name='PU-6203',
116129
)
117-
130+
118131
pu_taxonomy = {
119132
0: ('Normal', 'None'),
120133
1: ('Outer Race', 'Medium'),
121134
2: ('Inner Race', 'High'),
122135
}
123-
124-
train_data = Dataset.from_arrays(X_train, y_train, cond_train, fs=64000, physics=pu_physics, taxonomy=pu_taxonomy, name="PU_Train")
125-
calib_data = Dataset.from_arrays(X_calib, y_calib, cond_calib, fs=64000, physics=pu_physics, taxonomy=pu_taxonomy, name="PU_Calib")
126-
test_data = Dataset.from_arrays(X_test, y_test, cond_test, fs=64000, physics=pu_physics, taxonomy=pu_taxonomy, name="PU_Test")
127-
136+
137+
train_data = Dataset.from_arrays(
138+
X_train,
139+
y_train,
140+
cond_train,
141+
fs=64000,
142+
physics=pu_physics,
143+
taxonomy=pu_taxonomy,
144+
name='PU_Train',
145+
)
146+
calib_data = Dataset.from_arrays(
147+
X_calib,
148+
y_calib,
149+
cond_calib,
150+
fs=64000,
151+
physics=pu_physics,
152+
taxonomy=pu_taxonomy,
153+
name='PU_Calib',
154+
)
155+
test_data = Dataset.from_arrays(
156+
X_test,
157+
y_test,
158+
cond_test,
159+
fs=64000,
160+
physics=pu_physics,
161+
taxonomy=pu_taxonomy,
162+
name='PU_Test',
163+
)
164+
128165
model = CNSD()
129-
130-
print("\n[1] Training Neural Network on 900 RPM Data...")
166+
167+
print('\n[1] Training Neural Network on 900 RPM Data...')
131168
model.fit(train_data, epochs=20)
132-
133-
print("\n[2] Calibrating Tau threshold on 1500 RPM Data...")
169+
170+
print('\n[2] Calibrating Tau threshold on 1500 RPM Data...')
134171
taus = np.arange(1.0, 4.1, 0.5)
135172
best_gap = -np.inf
136173
best_tau = 1.0
137-
174+
138175
for tau in taus:
139176
model.symbolic.tau = float(tau)
140177
report = model.diagnose(calib_data)
141-
178+
142179
hb = headline_accuracy_by_verdict(report, y_calib)
143-
180+
144181
conf_acc = hb.get('CONFIRMED', {}).get('cnn_accuracy', 0.0)
145182
cnfl_acc = hb.get('CONFLICT', {}).get('cnn_accuracy', 0.0)
146183
gap = conf_acc - cnfl_acc if 'CONFIRMED' in hb and 'CONFLICT' in hb else 0.0
147-
148-
print(f"Calib tau={tau:.1f} | Conf={conf_acc:.3f} | Cnfl={cnfl_acc:.3f} | Gap={gap:+.3f}")
184+
185+
print(f'Calib tau={tau:.1f} | Conf={conf_acc:.3f} | Cnfl={cnfl_acc:.3f} | Gap={gap:+.3f}')
149186
if gap > best_gap:
150187
best_gap = gap
151188
best_tau = float(tau)
152-
153-
print(f"\n=> Selected optimal tau: {best_tau}")
189+
190+
print(f'\n=> Selected optimal tau: {best_tau}')
154191
model.symbolic.tau = best_tau
155-
156-
print("\n[3] Evaluating on Test Set (1500 RPM)...")
192+
193+
print('\n[3] Evaluating on Test Set (1500 RPM)...')
157194
report = model.diagnose(test_data)
158-
195+
159196
hb = headline_accuracy_by_verdict(report, y_test)
160-
197+
161198
# Calculate baseline CNN accuracy
162199
pred = np.array([r['predicted_class'] for r in report.records])
163200
baseline_acc = float((pred == np.asarray(y_test)).mean())
164-
165-
print("\n--- FINAL TEST RESULTS (CROSS-DOMAIN PU) ---")
166-
print(f"Baseline CNN Acc: {baseline_acc:.3f}")
201+
202+
print('\n--- FINAL TEST RESULTS (CROSS-DOMAIN PU) ---')
203+
print(f'Baseline CNN Acc: {baseline_acc:.3f}')
167204
if 'CONFIRMED' in hb:
168-
print(f"Physics-Confirmed Acc: {hb['CONFIRMED']['cnn_accuracy']:.3f} (n={hb['CONFIRMED']['n']})")
205+
print(
206+
f'Physics-Confirmed Acc: {hb["CONFIRMED"]["cnn_accuracy"]:.3f} (n={hb["CONFIRMED"]["n"]})'
207+
)
169208
if 'CONFLICT' in hb:
170-
print(f"Physics-Conflict Acc: {hb['CONFLICT']['cnn_accuracy']:.3f} (n={hb['CONFLICT']['n']})")
209+
print(
210+
f'Physics-Conflict Acc: {hb["CONFLICT"]["cnn_accuracy"]:.3f} (n={hb["CONFLICT"]["n"]})'
211+
)
171212
if 'INCONCLUSIVE' in hb:
172-
print(f"Physics-Inconclusive Acc:{hb['INCONCLUSIVE']['cnn_accuracy']:.3f} (n={hb['INCONCLUSIVE']['n']})")
173-
213+
print(
214+
f'Physics-Inconclusive Acc:{hb["INCONCLUSIVE"]["cnn_accuracy"]:.3f} (n={hb["INCONCLUSIVE"]["n"]})'
215+
)
216+
174217
if 'CONFIRMED' in hb and 'CONFLICT' in hb:
175218
gap = hb['CONFIRMED']['cnn_accuracy'] - hb['CONFLICT']['cnn_accuracy']
176-
print(f"GAP (CONF - CNFL): {gap:+.3f}")
177-
print("--------------------------------------------")
219+
print(f'GAP (CONF - CNFL): {gap:+.3f}')
220+
print('--------------------------------------------')

0 commit comments

Comments
 (0)