Приложение г. Эксперимент 4.

# Эксперимент 4

model_4 = Sequential()

model_4.add(Dense(128,input_dim=num_pixels, activation='relu'))

model_4.add(Dense(64,input_dim=num_pixels, activation='relu'))

model_4.add(Dense(32,input_dim=num_pixels, activation='relu'))

model_4.add(Dense(10, activation='softmax'))

model_4.compile(loss='categorical_crossentropy',optimizer='adam', metrics=['accuracy'])

history_4, scores_4  = make_history(model_4, 40)

y_pred_4 = prediction(model_4)

add_model_results(4,"DNN","D-128, D-64, D-32",4,"ReLU", scores_4[1], scores_4[0])

network_report(y_pred_4)

plot_loss(history_4)

Приложение д. Эксперимент 5.

model_5 = Sequential()

model_5.add(Conv2D(32, kernel_size=(3, 3),

                 activation='relu',

                 input_shape=(28, 28, 1)))

model_5.add(MaxPooling2D(pool_size=(2, 2)))

model_5.add(Dropout(0.25))

model_5.add(Flatten())

model_5.add(Dense(128, activation='relu'))

model_5.add(Dense(10, activation='softmax'))

model_5.summary()

model_5.compile(loss='categorical_crossentropy',optimizer='adam',metrics=['accuracy'])

# Загрузка данных

(X_train, y_train), (X_test, y_test) = mnist.load_data()

# Обработка

X_train = X_train / 255

X_test = X_test / 255

X_train = X_train.reshape(X_train.shape[0], 28, 28, 1)

X_test = X_test.reshape(X_test.shape[0], 28, 28, 1)

y_train = to_categorical(y_train)

y_test = to_categorical(y_test)

num_classes = y_test.shape[1]

history_5, scores_5 = make_history(model_5, 40)

y_pred_5 = prediction(model_5)

add_model_results(5,"CNN","Conv-32, MaxPooling, Dropout, Flatten, D-128",6,"ReLU", scores_5[1], scores_5[0])

network_report(y_pred_5)

plot_loss(history_5)

Приложение е. Эксперимент 6 и сравнение результатов.

model_6 = Sequential()

# Входной сверточнй слой

model_6.add(Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=(28, 28, 1)))

model_6.add(MaxPooling2D(pool_size=(2, 2)))

# Второй сверточный слой

model_6.add(Conv2D(64, kernel_size=(3, 3), activation='relu'))

model_6.add(MaxPooling2D(pool_size=(2, 2)))

# Третий сверточный слой

model_6.add(Conv2D(128, kernel_size=(3, 3), activation='relu'))

model_6.add(MaxPooling2D(pool_size=(2, 2)))

model_6.add(Flatten())

model_6.add(Dense(128, activation='relu'))

# Слой исключения

model_6.add(Dropout(0.25))

# Выходной слой

model_6.add(Dense(10, activation='softmax'))

# Характеристики модели

model_6.summary()

model_6.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

history_6, scores_6 = make_history(model_6, 40)

y_pred_6 = prediction(model_6)

add_model_results(6,"CNN","Conv-32,MaxPooling, Conv-64, MaxPooling, Conv-128, MaxPooling, Flatten, D-128, Dropout",10,"ReLU", scores_6[1], scores_6[0])

network_report(y_pred_6)

plot_loss(history_6)

results_df.sort_values(by=['Точность'], ascending=False)

from PIL import Image

from keras.models import load_model

# Обработка изображений

def preprocess_image(image_path):

    img = Image.open(image_path).convert('L')

    plt.imshow(img, cmap=plt.get_cmap('gray'))

    plt.show()

    # Преобразование в 28 на 28 пикселей

    img = img.resize((28, 28))

    img_array = np.array(img) / 255.0

    img_array = img_array.reshape(1, 28, 28, 1)

    return img_array

# Загрузка файлов

img1 = preprocess_image('/content/drive/MyDrive/my_digits/seven.png')

img2 = preprocess_image('/content/drive/MyDrive/my_digits/eight.png')

# Выполнение предсказаний

prediction_seven = model_5.predict(img1)

prediction_eight = model_5.predict(img2)

# Вывод предсказаний в удобном формате

predicted_digit_seven = np.argmax(prediction_seven)

predicted_digit_eight = np.argmax(prediction_eight)

print("Предсказание для семерки:", predicted_digit_seven)

print("Предсказание для восьмерки:", predicted_digit_eight)