from tensorflow.keras import Sequential
from tensorflow.keras.layers import Dense, Flatten, Conv2D

model = Sequential([
    Conv2D(32, 3, activation='relu', input_shape=(28, 28, 1)),
    Flatten(),
    Dense(128, activation='relu'),
    Dense(10, activation='softmax')
])

from tensorflow.keras import Model, Input
from tensorflow.keras.layers import Dense, Concatenate

inputs = Input(shape=(784,))
x = Dense(64, activation='relu')(inputs)
x = Dense(32, activation='relu')(x)
outputs = Dense(10, activation='softmax')(x)
model = Model(inputs=inputs, outputs=outputs)

from tensorflow.keras import Model
from tensorflow.keras.layers import Dense

class MyModel(Model):
    def __init__(self):
        super(MyModel, self).__init__()
        self.dense1 = Dense(64, activation='relu')
        self.dense2 = Dense(10, activation='softmax')
    
    def call(self, inputs):
        x = self.dense1(inputs)
        return self.dense2(x)

from tensorflow.keras import layers

# 常用層類型
# - Dense: 全連接層
# - Conv2D/Conv1D/Conv3D: 卷積層
# - LSTM/GRU/SimpleRNN: 循環(huán)層
# - Dropout: 丟棄層
# - BatchNormalization: 批量歸一化
# - Embedding: 嵌入層
# - MaxPooling2D/AveragePooling2D: 池化層
# - LayerNormalization: 層歸一化

from tensorflow.keras import losses

# 常用損失函數(shù)
# - BinaryCrossentropy: 二分類交叉熵
# - CategoricalCrossentropy: 多分類交叉熵
# - MeanSquaredError: 均方誤差
# - MeanAbsoluteError: 平均絕對(duì)誤差
# - Huber: Huber損失（回歸問(wèn)題）
# - SparseCategoricalCrossentropy: 稀疏多分類交叉熵

from tensorflow.keras import optimizers

# 常用優(yōu)化器
# - SGD: 隨機(jī)梯度下降（可帶動(dòng)量）
# - Adam: 自適應(yīng)矩估計(jì)
# - RMSprop: 均方根傳播
# - Adagrad: 自適應(yīng)梯度
# - Nadam: Nesterov Adam

from tensorflow.keras import metrics

# 常用指標(biāo)
# - Accuracy: 準(zhǔn)確率
# - Precision: 精確率
# - Recall: 召回率
# - AUC: ROC曲線下面積
# - MeanSquaredError: 均方誤差
# - MeanAbsoluteError: 平均絕對(duì)誤差

from tensorflow.keras import callbacks

# 常用回調(diào)
# - ModelCheckpoint: 模型保存
# - EarlyStopping: 早停
# - TensorBoard: TensorBoard可視化
# - ReduceLROnPlateau: 動(dòng)態(tài)調(diào)整學(xué)習(xí)率
# - CSVLogger: 訓(xùn)練日志記錄

from tensorflow.keras.preprocessing import image, text, sequence

# 圖像預(yù)處理
# - ImageDataGenerator: 圖像增強(qiáng)（TF 2.x 風(fēng)格）
# - load_img, img_to_array: 圖像加載轉(zhuǎn)換

# 文本預(yù)處理
# - Tokenizer: 文本分詞
# - pad_sequences: 序列填充

from tensorflow.keras.applications import (
    VGG16, ResNet50, MobileNet,
    InceptionV3, EfficientNetB0
)

# 加載預(yù)訓(xùn)練模型
base_model = ResNet50(weights='imagenet', include_top=False)

from tensorflow.keras import utils

# 常用工具
# - to_categorical: 類別編碼
# - plot_model: 模型結(jié)構(gòu)可視化
# - normalize: 數(shù)據(jù)標(biāo)準(zhǔn)化

import tensorflow as tf
from tensorflow.keras import layers, models

# 1. 數(shù)據(jù)準(zhǔn)備
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
x_train = x_train.reshape(-1, 28, 28, 1).astype('float32') / 255.0
x_test = x_test.reshape(-1, 28, 28, 1).astype('float32') / 255.0

# 2. 構(gòu)建模型
model = models.Sequential([
    layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)),
    layers.MaxPooling2D((2, 2)),
    layers.Conv2D(64, (3, 3), activation='relu'),
    layers.MaxPooling2D((2, 2)),
    layers.Flatten(),
    layers.Dense(128, activation='relu'),
    layers.Dropout(0.5),
    layers.Dense(10, activation='softmax')
])

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

# 4. 訓(xùn)練模型
history = model.fit(
    x_train, y_train,
    epochs=10,
    batch_size=32,
    validation_split=0.2,
    callbacks=[
        tf.keras.callbacks.EarlyStopping(patience=3),
        tf.keras.callbacks.ModelCheckpoint('best_model.h5')
    ]
)

# 5. 評(píng)估模型
test_loss, test_acc = model.evaluate(x_test, y_test)

# 6. 使用模型預(yù)測(cè)
predictions = model.predict(x_test[:5])

from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.preprocessing.sequence import pad_sequences

# 1. 文本預(yù)處理
tokenizer = Tokenizer(num_words=10000)
tokenizer.fit_on_texts(texts)
sequences = tokenizer.texts_to_sequences(texts)
padded_sequences = pad_sequences(sequences, maxlen=200)

# 2. 構(gòu)建文本分類模型
model = models.Sequential([
    layers.Embedding(10000, 128, input_length=200),
    layers.Bidirectional(layers.LSTM(64, return_sequences=True)),
    layers.GlobalMaxPooling1D(),
    layers.Dense(64, activation='relu'),
    layers.Dense(1, activation='sigmoid')  # 二分類
])

class CustomLayer(layers.Layer):
    def __init__(self, units=32):
        super(CustomLayer, self).__init__()
        self.units = units
    
    def build(self, input_shape):
        self.w = self.add_weight(
            shape=(input_shape[-1], self.units),
            initializer='random_normal',
            trainable=True
        )
        self.b = self.add_weight(
            shape=(self.units,),
            initializer='zeros',
            trainable=True
        )
    
    def call(self, inputs):
        return tf.matmul(inputs, self.w) + self.b

def custom_loss(y_true, y_pred):
    mse = tf.keras.losses.mean_squared_error(y_true, y_pred)
    penalty = tf.reduce_mean(tf.square(y_pred))
    return mse + 0.01 * penalty

# 多輸入
input1 = Input(shape=(64,))
input2 = Input(shape=(128,))

# 多輸出
output1 = Dense(1, name='regression')(merged)
output2 = Dense(5, activation='softmax', name='classification')(merged)

model = Model(inputs=[input1, input2], outputs=[output1, output2])

# 保存整個(gè)模型
model.save('my_model.h5')

# 保存為SavedModel格式（用于TF Serving）
model.save('my_model', save_format='tf')

# 轉(zhuǎn)換為TensorFlow Lite（移動(dòng)端）
converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()