import numpy as np
import tensorflow as tf
# Load Data
data_path = "../data/mixedtype-wafer-defect-datasets/Wafer_Map_Datasets.npz"
data = np.load(data_path)
images = data["arr_0"]
labels = data["arr_1"]
print("✓ Data loaded")
# Load the Phase 2 models
MODELS_DIR = "phase2_models"
model_p2_cnn = tf.keras.models.load_model(f"{MODELS_DIR}/model_p2_cnn.keras")
model_p2_tl = tf.keras.models.load_model(f"{MODELS_DIR}/model_p2_tl.keras")
print("✓ Models loaded")✓ Data loaded
✓ Models loaded# Default pixels with 3 to 0
images[images == 3] = 0
# String Label Mapping
label_mapping = {
"00000000": "Normal",
"10000000": "Center",
"01000000": "Donut",
"00100000": "Edge_Loc",
"00010000": "Edge_Ring",
"00001000": "Loc",
"00000100": "Near_Full",
"00000010": "Scratch",
"00000001": "Random",
"10100000": "Center+Edge_Loc",
"10010000": "Center+Edge_Ring",
"10001000": "Center+Loc",
"10000010": "Center+Scratch",
"01100000": "Donut+Edge_Loc",
"01010000": "Donut+Edge_Ring",
"01001000": "Donut+Loc",
"01000010": "Donut+Scratch",
"00101000": "Edge_Loc+Loc",
"00100010": "Edge_Loc+Scratch",
"00011000": "Edge_Ring+Loc",
"00010010": "Edge_Ring+Scratch",
"00001010": "Loc+Scratch",
"10101000": "Center+Edge_Loc+Loc",
"10100010": "Center+Edge_Loc+Scratch",
"10011000": "Center+Edge_Ring+Loc",
"10010010": "Center+Edge_Ring+Scratch",
"10001010": "Center+Loc+Scratch",
"01101000": "Donut+Edge_Loc+Loc",
"01100010": "Donut+Edge_Loc+Scratch",
"01011000": "Donut+Edge_Ring+Loc",
"01010010": "Donut+Edge_Ring+Scratch",
"01001010": "Donut+Loc+Scratch",
"00101010": "Edge_Loc+Loc+Scratch",
"00011010": "Edge_Ring+Loc+Scratch",
"10101010": "Center+Edge_Loc+Loc+Scratch",
"10011010": "Center+Edge_Ring+Loc+Scratch",
"01101010": "Donut+Edge_Loc+Loc+Scratch",
"01011010": "Donut+Edge_Ring+Loc+Scratch",
}
unique_patterns = sorted(label_mapping.keys())
pattern_to_id = {pattern: idx for idx, pattern in enumerate(unique_patterns)}
id_to_pattern = {idx: label_mapping[pattern] for pattern, idx in pattern_to_id.items()}
label_str_arr = np.array(["".join(map(str, map(int, row))) for row in labels])
X_test = np.eye(3, dtype=np.float32)[images.astype(int)] # Shape: (52, 52, 3)
y_class_test = np.array([pattern_to_id[s] for s in label_str_arr])# Pick a random wafer map from test set
random_idx = np.random.randint(0, len(X_test))
random_wafer = X_test[random_idx : random_idx + 1]
true_class_id = y_class_test[random_idx]
true_pattern_name = id_to_pattern[true_class_id]
print(f"Random wafer shape: {random_wafer.shape}")
print(f"True defect pattern: {true_pattern_name}")
# Make predictions
pred_cnn = model_p2_cnn.predict(random_wafer, verbose=0) # Shape: (1, 38)
pred_tl = model_p2_tl.predict(random_wafer, verbose=0) # Shape: (1, 38)
# Get predicted class IDs (argmax)
pred_class_cnn = pred_cnn[0].argmax()
pred_class_tl = pred_tl[0].argmax()
# Get predicted pattern names
pred_pattern_cnn = id_to_pattern[pred_class_cnn]
pred_pattern_tl = id_to_pattern[pred_class_tl]
# Get confidence scores
conf_cnn = pred_cnn[0, pred_class_cnn]
conf_tl = pred_tl[0, pred_class_tl]
print("\nPredictions:")
print(f" Custom CNN: {pred_pattern_cnn} (confidence: {conf_cnn:.2%})")
print(f" MobileNetV2: {pred_pattern_tl} (confidence: {conf_tl:.2%})")
print(f" Match: {'✓' if pred_class_cnn == true_class_id else '✗'} (Custom CNN)")
print(f" Match: {'✓' if pred_class_tl == true_class_id else '✗'} (MobileNetV2)")Random wafer shape: (1, 52, 52, 3)
True defect pattern: Loc+Scratch
Predictions:
Custom CNN: Loc+Scratch (confidence: 99.89%)
MobileNetV2: Loc+Scratch (confidence: 98.48%)
Match: ✓ (Custom CNN)
Match: ✓ (MobileNetV2)