设置tf.变量Python/TensorFlow的初始值

我有这个功能：

def new_weights(shape):
    return tf.Variable(tf.truncated_normal(shape, stddev=0.05))

我这样称呼它，例如：

# shape = [filter_size, filter_size, num_filters, num_input_channels]
shape = [1, 1, 8, 1]

weights = new_weights(shape)

我想用以下值初始化权重：

weights = [1, 2, 3, 4, 5, 6, 7, 8]

在用这些值初始化它之后，我希望它被更新（可训练）

---

如何执行此操作？

您可以使用分配功能

shape = [1, 1, 8, 1]

weights = new_weights(shape)

ws = [1, 2, 3, 4, 5, 6, 7, 8]

ws = np.array(ws).reshape(shape)
weights = weights.assign(ws)

---

我认为您可以使用如下函数：

def new_weights(shape):
    total = np.prod(shape)
    init_data = np.array(range(1, 1+ total)).reshape(shape)
    return tf.get_variable(name='weights', 
                           initializer = tf.constant_initializer(init_data), 
                           shape = shape)

并检查它：

shape = [1, 1, 8, 1]
weights = new_weights(shape)
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print(sess.run(weights))

---

关于tf.Variable（初始值=权重，…）？

这并不是问题的答案。若要评论或要求作者澄清，请在其帖子下方留下评论-