NameError: name 'model' is not defined. Did you mean: 'Model'?

这个错误提示表明你的代码中引用了一个未定义的变量 model。看起来是因为在定义模型的最后部分代码中缺少了括号，导致 model 变量没有被正确定义。

以下是修正后的代码：

import tensorflow as tf
from tensorflow.keras.layers import Layer, Conv2D, Dense, Input, Reshape
from tensorflow.keras.models import Model
from tensorflow.keras.datasets import mnist
import numpy as np
import matplotlib.pyplot as plt

# 加载并准备MNIST数据集
(train_images, train_labels), (test_images, test_labels) = mnist.load_data()
train_images = train_images.reshape((60000, 28, 28, 1)).astype('float32') / 255
test_images = test_images.reshape((10000, 28, 28, 1)).astype('float32') / 255

# 定义胶囊层
class CapsuleLayer(Layer):
    def __init__(self, num_capsules, dim_capsule, routings, **kwargs):
        super(CapsuleLayer, self).__init__(**kwargs)
        self.num_capsules = num_capsules
        self.dim_capsule = dim_capsule
        self.routings = routings

    def build(self, input_shape):
        # input_shape: (batch_size, height, width, channels)
        self.kernel = self.add_weight(
            shape=[input_shape[-1], self.num_capsules * self.dim_capsule],
            initializer='glorot_uniform',
            trainable=True,
            name='capsule_kernel'
        )

    def call(self, inputs):
        # inputs: (batch_size, height, width, channels)
        # Reshape inputs to (batch_size, height * width, channels)
        x = tf.reshape(inputs, (tf.shape(inputs)[0], -1, inputs.shape[-1]))

        # Calculate u_hat
        u_hat = tf.matmul(x, self.kernel)
        u_hat = tf.reshape(u_hat, (tf.shape(u_hat)[0], -1, self.num_capsules, self.dim_capsule))

        # Initialize the Logits (b) as zeros
        b = tf.zeros((tf.shape(inputs)[0], self.num_capsules), dtype=tf.float32)

        # Routing algorithm
        for i in range(self.routings):
            c = tf.nn.softmax(b, axis=1)  # Softmax over the capsules
            s = tf.matmul(tf.reshape(c, (tf.shape(c)[0], -1, 1)), u_hat, transpose_b=True)  # Weighted sum of u_hat vectors
            s = tf.squeeze(s, 2)  # Remove the singleton dimension
            v = tf.nn.squash(s)  # Squash the vector
            b = b + tf.reduce_sum(v, axis=2)  # Update the logits

        # Squash the final vectors
        v = tf.nn.squash(b[..., tf.newaxis])

        return tf.reshape(v, (tf.shape(inputs)[0], -1, self.num_capsules, self.dim_capsule))

    def compute_output_shape(self, input_shape):
        return (input_shape[0], input_shape[1] * input_shape[2], self.num_capsules, self.dim_capsule)


# 构建胶囊网络模型
inputs = Input(shape=(28, 28, 1))
x = Conv2D(256, kernel_size=9, strides=2, padding='valid')(inputs)
x = tf.keras.layers.BatchNormalization()(x)
x = tf.keras.layers.LeakyReLU()(x)

capsule_layer = CapsuleLayer(num_capsules=10, dim_capsule=16, routings=3)
primarycaps = capsule_layer(x)

# 将输出维度调整为 (batch_size, 10)
primarycaps = Reshape((-1, 10))(primarycaps)

outputs = Dense(10, activation='softmax')(primarycaps)

model = Model(inputs=inputs, outputs=outputs)  # 确保定义了 model 变量

# 定义损失函数
def margin_loss(y_true, y_pred):
    L = y_true * tf.square(tf.maximum(0., 0.9 - y_pred)) + 0.5 * (1 - y_true) * tf.square(tf.maximum(0., y_pred - 0.1))
    return tf.reduce_mean(tf.reduce_sum(L, axis=[1, 2]))

model.compile(optimizer='adam', loss=margin_loss, metrics=['accuracy'])
model.summary()

# 训练模型
history = model.fit(train_images, train_labels, epochs=50, validation_data=(test_images, test_labels), batch_size=100)

# 如果需要，还可以绘制训练和验证的损失及准确率曲线
plt.plot