import numpy as np
from keras import Sequential
from keras.src.layers import Dense, Input
from tensorflow.python.keras.utils.np_utils import to_categorical

# 1️⃣ 입력 데이터 (랜덤 생성)
x_data = np.random.random((10, 4))              # 10개 샘플, 4개의 피처
y_labels = np.random.randint(0,3,size=(10,))    # 0~2 클래스 중 랜덤 레이블
print(f'x_data: {x_data}')
print(f'y_labels: {y_labels}')

# 2️⃣ 원-핫 인코딩 (to_categorical)
# → 클래스 0,1,2를 [1,0,0], [0,1,0], [0,0,1] 형태로 변환
y_data = to_categorical(y_labels, num_classes=3)
print(f'y_data: {y_data}')
print()

# 3️⃣ 모델 구성: 입력(4) → 은닉(30, ReLU) → 출력(3, Softmax)
model = Sequential([
    Dense(30, activation='relu', input_shape=(4,)),
    Dense(3, activation='softmax')
])

# 4️⃣ 모델 설정 및 학습
# 손실함수: categorical_crossentropy (다중분류용)
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
model.fit(x_data, y_data, epochs=10, validation_split=0.3, verbose=2)

# accuracy : 훈련 정확도 / val_accuracy : 검증 정확도

from keras.src.datasets import mnist
from matplotlib import pyplot as plt

# 1️⃣ 데이터 로드
(x_train, y_train), (x_test, y_test) = mnist.load_data()

print(x_train.shape)  # (60000, 28, 28)
print(y_train.shape)  # (60000,)
print(x_test.shape)   # (10000, 28, 28)
print(y_test.shape)   # (10000,)

print(x_train[0], y_train[0])  # 첫 번째 이미지와 정답 출력

# 2️⃣ 이미지 시각화(2×3 그리드)
nrow, ncol = 2, 3
fig, axes = plt.subplots(nrow, ncol, figsize=[10, 6])

for idx in range(nrow * ncol):
    ax = axes[idx // ncol, idx % ncol]
    ax.imshow(x_train[idx], cmap='gray')
    ax.axis('off')
    ax.set_title(f'label: {y_train[idx]}')

plt.show()

설명

• mnist.load_data(): 자동으로 훈련/테스트 분리
• x_train: 이미지 (28×28), y_train: 라벨(0~9)
• 간단한 시각화로 데이터 감 잡기

Ex03_mnist.py — MLP로 MNIST 다중분류

import numpy as np
from keras import Sequential
from keras.src.datasets import mnist
from keras.src.layers import Dense
from keras.src.utils import to_categorical
from matplotlib import pyplot as plt

# 1) 데이터 적재
(x_train, y_train), (x_test, y_test) = mnist.load_data()

# 2) 입력 차원 계산(28*28=784)
x_column = x_train.shape[1] * x_train.shape[2]
print('x_column:', x_column)

# 3) 평탄화 + 정규화
x_train = x_train.reshape(x_train.shape[0], x_column).astype('float32')/255.0
x_test  = x_test.reshape(x_test.shape[0],  x_column).astype('float32')/255.0

# 4) 원-핫 라벨
CLASSES = 10
y_train = to_categorical(y_train, num_classes=CLASSES)
y_test  = to_categorical(y_test,  num_classes=CLASSES)

# 5) 모델
model = Sequential()
model.add(Dense(units=512, activation="relu", input_shape=(x_column,)))
model.add(Dense(units=CLASSES, activation="softmax"))

# 6) 컴파일 & 학습
model.compile(loss="categorical_crossentropy", optimizer="rmsprop", metrics=["accuracy"])
model.fit(x_train, y_train, epochs=10, batch_size=64, validation_split=0.2, verbose=1)

# 7) 평가 & 예측
score = model.evaluate(x_test, y_test)
print(f'score : {score}')  # [loss, acc]

prediction = model.predict(x_test)
predict_class = np.argmax(prediction[0])
print(f'prediction class : {predict_class}')

Ex04_mnist.py — Fashion-MNIST 분류 & 시각화

import numpy as np
from keras import Sequential
from keras.src.datasets import fashion_mnist
from keras.src.layers import Dense
from keras.src.utils import to_categorical
from matplotlib import pyplot as plt

# 1) 데이터 적재(Fashion-MNIST)
(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()
print(x_train.shape, y_train.shape, x_test.shape, y_test.shape)

# 2) 평탄화 + 정규화
x_column = x_train.shape[1] * x_train.shape[2]
x_train = x_train.reshape(x_train.shape[0], x_column).astype('float32')/255.0
x_test  = x_test.reshape(x_test.shape[0],  x_column).astype('float32')/255.0

# 3) 클래스 라벨
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
               'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']

# 4) 원-핫 라벨
CLASSES = 10
y_train = to_categorical(y_train, num_classes=CLASSES)
y_test  = to_categorical(y_test,  num_classes=CLASSES)

# 5) 모델: 128-128 ReLU → 10 Softmax
model = Sequential()
model.add(Dense(units=128, activation='relu', input_shape=(x_column,)))
model.add(Dense(units=128, activation='relu'))
model.add(Dense(units=CLASSES, activation='softmax'))

# 6) 컴파일 & 학습
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.fit(x_train, y_train, batch_size=10, epochs=10, validation_data=(x_test, y_test))

# 7) 예측 & 평가
prediction = model.predict(x_test)
score = model.evaluate(x_test, y_test)
print(f'손실률 : {score[0]}, 정확도 {score[1]}')

# (권장) 개별 샘플의 argmax
print('예측')
print('y_test', np.argmax(y_test[0]))
print('y_pred', np.argmax(prediction[0]))

# 8) 4x4 그리드 시각화
nrow, ncol = 4, 4
fig, axes = plt.subplots(nrow, ncol, figsize=[10, 6])
for idx in range(nrow * ncol):
    ax = axes[idx//ncol, idx%ncol]
    ax.imshow(x_train[idx].reshape(28,28))
    ax.axis('off')
    ax.set_title(f'label: {class_names[np.argmax(y_train[idx])]}')
plt.show()