Tensorfow初始化一定範圍僅

Question

他那裏，

我有一個關於在該變量的作用域被初始化，或至少控制，該變量的作用域是在運行期間使用的問題。

採取例如這個簡單的一段代碼

import numpy as np 
import tensorflow as tf 

with tf.variable_scope('0') as scope: 
    place_holder_batch_x = tf.Variable(np.random.rand(11,6), dtype=tf.float64) 
    place_holder_batch_y = tf.Variable(np.random.rand(8,5), dtype=tf.float64) 
    rnn_cell = tf.nn.rnn_cell.BasicRNNCell(3) 
    z = place_holder_batch_x*2 

with tf.variable_scope('1') as scope: 
    place_holder_batch_x = tf.Variable(np.random.rand(10,5), dtype=tf.float64) 
    place_holder_batch_y = tf.Variable(np.random.rand(9,6), dtype=tf.float64) 
    rnn_cell = tf.nn.rnn_cell.BasicRNNCell(4) 
    z = place_holder_batch_x*2 

init = tf.global_variables_initializer() 

sess = tf.Session() 
sess.run(init) 
print(sess.run(z).shape) 

如果我會運行此按原樣我會得到如在可變範圍定義的變量z的形狀「1」。 但是如何指定在會話期間使用哪個變量範圍？我找不到在stackoverflow或在文檔中的任何答案...

當然，我可以只重命名z的z1和z2 ...但我想繼續留在這兩個範圍看起來很像對方並使用相同的名稱...

Answer 1

試試這個：

import numpy as np 
import tensorflow as tf 


g1 = tf.Graph() 
with g1.as_default() as g: 
    with tf.variable_scope('0') as scope: 
     place_holder_batch_x = tf.Variable(np.random.rand(11,6), dtype=tf.float64) 
     place_holder_batch_y = tf.Variable(np.random.rand(8,5), dtype=tf.float64) 
     rnn_cell = tf.nn.rnn_cell.BasicRNNCell(3) 
     z = place_holder_batch_x*2 
g2 = tf.Graph() 
with g2.as_default() as g: 
    with tf.variable_scope('1') as scope: 
     place_holder_batch_x = tf.Variable(np.random.rand(10,5), dtype=tf.float64) 
     place_holder_batch_y = tf.Variable(np.random.rand(9,6), dtype=tf.float64) 
     rnn_cell = tf.nn.rnn_cell.BasicRNNCell(4) 
     z = place_holder_batch_x*2 

tf.reset_graph_default() 

graph_to_be_used = g1 

with tf.session(graph = graph_to_be_used) as sess: 
    init = tf.global_variables_initializer() 

    sess.run(init) 
    print(sess.run(z).shape)