在TensorFlow中训练后没有更新变量，即使是在简单逻辑回归的均匀随机启动时也是如此

import tensorflow as tf

# 生成模拟数据
X_train = tf.random.normal([100, 1])
y_train = tf.cast(X_train > 0, tf.float32)

# 定义模型参数
W = tf.Variable(tf.random.normal([1, 1]), name='weight')
b = tf.Variable(tf.zeros([1]), name='bias')

# 定义模型
def logistic_regression(x):
    return tf.sigmoid(tf.matmul(x, W) + b)

# 定义损失函数
def loss_fn(y_true, y_pred):
    return tf.reduce_mean(-y_true * tf.math.log(y_pred) - (1 - y_true) * tf.math.log(1 - y_pred))

# 定义优化器
optimizer = tf.optimizers.SGD(learning_rate=0.01)

# 训练操作
train_op = optimizer.minimize(loss_fn, var_list=[W, b])

# 初始化变量
init = tf.global_variables_initializer()

# 运行训练
with tf.compat.v1.Session() as sess:
    sess.run(init)
    for epoch in range(100):
        _, loss = sess.run([train_op, loss_fn(y_train, logistic_regression(X_train))])
        if epoch % 10 == 0:
            print(f'Epoch {epoch}, Loss: {loss}')