Upload New File

test_one.py

+# test_one.py 
+
+import numpy as np
+import os
+import sys
+import matplotlib.pyplot as plt
+from collections import defaultdict
+
+def get_working_directory(): #funkcja
+    if getattr(sys, 'frozen', False):
+        return os.path.dirname(sys.executable)
+    else:
+        return os.path.dirname(os.path.abspath(__file__))
+os.chdir(get_working_directory())
+
+#import funkcji z pliku model.py
+from model import (
+    conv_forward,pool_forward,relu,softmax
+)
+
+#definicja nazw klas dla lepszej czytelności
+CLASS_NAMES = [
+    "Airplane",
+    "Bird",
+    "Car",
+    "Cat",
+    "Deer",
+    "Dog",
+    "Horse",
+    "Monkey",
+    "Ship",
+    "Truck"
+]
+
+def read_stl10_images(path_to_bin, num_images):
+    #funkcja do wczytywania obrazów z pliku binarnego STL-10
+    with open(path_to_bin, 'rb') as f:
+        everything = np.fromfile(f, dtype=np.uint8)
+        images = everything.reshape(num_images, 3, 96, 96)
+        images = np.transpose(images, (0, 2, 3, 1)) 
+    return images
+
+def read_stl10_labels(path_to_bin, num_labels):
+    #funkcja do wczytywania etykiet z pliku binarnego STL-10
+    with open(path_to_bin, 'rb') as f:
+        labels = np.fromfile(f, dtype=np.uint8)
+    labels = labels - 1 
+    return labels
+
+
+def load_stl10(root_path="./stl10_binary"):
+    #funkcja do ładowania zestawu danych STL-10
+    X_train = read_stl10_images(os.path.join(root_path, 'train_X.bin'), 5000)
+    y_train = read_stl10_labels(os.path.join(root_path, 'train_y.bin'), 5000)
+    X_test = read_stl10_images(os.path.join(root_path, 'test_X.bin'), 8000)
+    y_test = read_stl10_labels(os.path.join(root_path, 'test_y.bin'), 8000)
+    return X_train, y_train, X_test, y_test
+
+#wczytywanie danych treningowych i testowych
+X_train, y_train, X_test, y_test = load_stl10()
+#normalizacja danych testowych
+X_test = X_test / 255.0
+
+checkpoint_path = "model_params.npz"  
+
+
+#wczytanie parametrów modelu z pliku .npz
+data = np.load(checkpoint_path)
+W1 = data["W1"]
+b1 = data["b1"]
+W2_conv = data["W2_conv"]
+b2_conv = data["b2_conv"]
+W_fc1 = data["W_fc1"]
+b_fc1 = data["b_fc1"]
+W_fc2 = data["W_fc2"]
+b_fc2 = data["b_fc2"]
+
+conv_hparams1 = {"stride":1, "pad":1}  
+pool_hparams1 = {"f":2, "stride":2}    
+conv_hparams2 = {"stride":1, "pad":1} 
+pool_hparams2 = {"f":2, "stride":2}    
+
+
+def forward_model(X, W1, b1, W2_conv, b2_conv, W_fc1, b_fc1, W_fc2, b_fc2):
+    #foward dla sieci
+    A1, _ = conv_forward(X, W1, b1, conv_hparams1, activation=relu)
+    A2, _ = pool_forward(A1, pool_hparams1, mode="max")
+    A3, _ = conv_forward(A2, W2_conv, b2_conv, conv_hparams2, activation=relu)
+    A4, _ = pool_forward(A3, pool_hparams2, mode="max")
+
+
+    A4_flat = A4.reshape(A4.shape[0], -1)
+    Z_fc1 = A4_flat.dot(W_fc1) + b_fc1
+    A_fc1 = relu(Z_fc1)
+    Z_fc2 = A_fc1.dot(W_fc2) + b_fc2
+    A_fc2 = softmax(Z_fc2)
+
+    return A_fc2
+
+
+print("Wczytano parametry z pliku:", checkpoint_path)
+
+
+while True:
+    index_str = input(f"Podaj indeks obrazu testowego (0 - {X_test.shape[0]-1}) lub 'q' aby wyjść: ")
+    if index_str.lower() == 'q':
+        break
+
+    try:
+        index = int(index_str)
+        if index < 0 or index >= X_test.shape[0]:
+            print(f"Nieprawidłowy indeks. Proszę podać liczbę z zakresu 0 - {X_test.shape[0]-1}.")
+            continue
+    except ValueError:
+        print("Proszę podać liczbę lub 'q'.")
+        continue
+
+    img = X_test[index]          
+    true_label = y_test[index]  
+
+    A_fc2_test = forward_model(img.reshape(1, 96, 96, 3), 
+                               W1, b1, 
+                               W2_conv, b2_conv, 
+                               W_fc1, b_fc1, 
+                               W_fc2, b_fc2)
+    pred_label = np.argmax(A_fc2_test, axis=1)[0]
+
+
+    correctness = "Poprawnie" if pred_label == true_label else "Niepoprawnie"
+
+    #wyświetlenie obrazu (obracanie o 90 stopni w lewo dla lepszej wizualizacji)
+    rotated_img = np.rot90(img, k=3)
+
+    plt.imshow(rotated_img)
+    plt.axis('off')  #wyłączenie osi dla lepszej estetyki
+    plt.title(f"Prawdziwa etykieta: {CLASS_NAMES[true_label]}, Predykcja: {CLASS_NAMES[pred_label]} ({correctness})")
+    plt.show()
+
+#koniec programu
+print("Program zakończył działanie.")