Fashion-MNIST影像分類別MLP模型建構

import numpy as np
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.datasets import fashion_mnist
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Flatten
from tensorflow.keras.utils import to_categorical
from sklearn.model_selection import train_test_split

# 載入 Fashion-MNIST 資料集
(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()

# 資料預處理：正規化和展平
x_train = x_train.astype('float32') / 255.0
x_test = x_test.astype('float32') / 255.0
x_train = x_train.reshape(-1, 784)
x_test = x_test.reshape(-1, 784)

# 標籤預處理：One-hot 編碼
y_train = to_categorical(y_train, num_classes=10)
y_test = to_categorical(y_test, num_classes=10)

# 資料集分割：訓練集和驗證集
x_train, x_val, y_train, y_val = train_test_split(x_train, y_train, test_size=0.2, random_state=42)

# 建立 MLP 模型
model = Sequential([
    Dense(128, activation='relu', input_shape=(784,)),
    Dense(64, activation='relu'),
    Dense(10, activation='softmax')
])

# 編譯模型
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

# 訓練模型
history = model.fit(x_train, y_train, epochs=10, batch_size=32, validation_data=(x_val, y_val))

# 評估模型
loss, accuracy = model.evaluate(x_test, y_test)
print(f"測試集損失：{loss:.4f}")
print(f"測試集準確率：{accuracy:.4f}")

# 提示
我想要為Fashion-MNIST資料集建立一個簡單的分類別模型（TAG 1.1），該模型由單層MLP組成（PIC 2.2）。它應包含以下步驟（TAG 1.2），請一步一步提供程式碼，並等待使用者的反饋（PIC 2.2）。
- 資料預處理：歸一化畫素值、將影像展平為向量、編碼分類別標籤。
- 資料分割：將資料集分成訓練集、驗證集和測試集。
- 模型選擇：選擇多層感知器（MLP）作為分類別模型。
- 模型訓練：在訓練資料上訓練MLP。
- 模型評估：使用準確率、精確度、召回率和混淆矩陣等指標評估模型的效能。

我是初學者（PIC 2.1），因此請提供逐步的程式碼，並嚴格遵循以下指令（TAG 1.3）：
- 程式碼應相容Jupyter Notebook。
- 確保每一行程式碼都有詳細的註解。
- 在提供程式碼之前，詳細解釋每一行程式碼，這些解釋將被複製到Notebook的文字區塊中。

# 步驟1：資料預處理
import tensorflow as tf
from tensorflow.keras.datasets import fashion_mnist
from tensorflow.keras.utils import to_categorical

# 載入Fashion-MNIST資料集
(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()

# 將畫素值歸一化到0到1之間
x_train = x_train.astype('float32') / 255
x_test = x_test.astype('float32') / 255

# 將影像展平為向量
x_train = x_train.reshape((-1, 784))
x_test = x_test.reshape((-1, 784))

# 對標籤進行one-hot編碼
y_train = to_categorical(y_train, 10)
y_test = to_categorical(y_test, 10)

#### 內容解密：
1. `fashion_mnist.load_data()` 載入Fashion-MNIST資料集，包括訓練和測試資料及其標籤。
2. `astype('float32') / 255` 將畫素值從0-255歸一化到0-1之間，有助於模型的收斂。
3. `reshape((-1, 784))` 將28x28的影像展平為784維的向量，因為MLP需要一維輸入。
4. `to_categorical(y_train, 10)` 對標籤進行one-hot編碼，使其適合多分類別問題。

import numpy as np
from tensorflow.keras.datasets import fashion_mnist
from sklearn.preprocessing import LabelEncoder
from sklearn.model_selection import train_test_split

# 載入Fashion-MNIST資料集
(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()

# 正規化畫素值
def normalize_images(train_data, test_data):
    train_data = train_data.astype('float32')
    test_data = test_data.astype('float32')
    train_data /= 255.0
    test_data /= 255.0
    return train_data, test_data

# 將影像展平為向量
def flatten_images(train_data, test_data):
    train_data = train_data.reshape(train_data.shape[0], -1)
    test_data = test_data.reshape(test_data.shape[0], -1)
    return train_data, test_data

# 編碼分類別標籤
def encode_labels(train_labels, test_labels):
    label_encoder = LabelEncoder()
    train_labels = label_encoder.fit_transform(train_labels)
    test_labels = label_encoder.transform(test_labels)
    return train_labels, test_labels

# 套用資料預處理步驟
x_train, x_test = normalize_images(x_train, x_test)
x_train, x_test = flatten_images(x_train, x_test)
y_train, y_test = encode_labels(y_train, y_test)

# 列印預處理後的資料形狀
print("x_train shape:", x_train.shape)
print("y_train shape:", y_train.shape)
print("x_test shape:", x_test.shape)
print("y_test shape:", y_test.shape)

# 分割資料集為訓練、驗證和測試集
def split_dataset(train_data, train_labels, test_data, test_labels, validation_size=0.1, random_state=42):
    x_train, x_val, y_train, y_val = train_test_split(train_data, train_labels, test_size=validation_size, random_state=random_state)
    # 注意：此處應當直接傳回x_train, y_train, x_val, y_val和原始的x_test, y_test，而不是再次分割x_test和y_test
    return x_train, y_train, x_val, y_val, test_data, test_labels

# 套用資料分割
x_train, y_train, x_val, y_val, x_test, y_test = split_dataset(x_train, y_train, x_test, y_test)

from sklearn.model_selection import train_test_split

def split_dataset(x_train, y_train, x_test, y_test, val_size=0.2, test_size=0.2, random_state=42):
    # 將訓練集進一步分割為訓練集和驗證集
    x_train, x_val, y_train, y_val = train_test_split(x_train, y_train, test_size=val_size, random_state=random_state)
    return x_train, x_val, y_train, y_val

# 分割資料集
x_train, x_val, y_train, y_val = split_dataset(x_train, y_train, x_test, y_test)

# 輸出分割後資料集的形狀以進行驗證
print("訓練集：")
print("x_train 形狀：", x_train.shape)
print("y_train 形狀：", y_train.shape)
print("\n驗證集：")
print("x_val 形狀：", x_val.shape)
print("y_val 形狀：", y_val.shape)
print("\n測試集：")
print("x_test 形狀：", x_test.shape)
print("y_test 形狀：", y_test.shape)

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense

def create_single_layer_mlp(input_shape, num_classes):
    model = Sequential()
    # 新增單一隱藏層，包含 128 個神經元，並使用 ReLU 作為啟用函式
    model.add(Dense(128, activation='relu', input_shape=input_shape))
    # 新增輸出層，使用 softmax 啟用函式進行多類別分類別
    model.add(Dense(num_classes, activation='softmax'))
    return model

# 定義輸入形狀和類別數量
input_shape = x_train.shape[1:]
num_classes = len(np.unique(y_train))

# 建立單層 MLP 模型
model = create_single_layer_mlp(input_shape, num_classes)

# 輸出模型摘要
model.summary()

# 編譯模型
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

# 訓練模型
history = model.fit(x_train, y_train, epochs=10, batch_size=128, validation_data=(x_val, y_val))