python - How to save a specific variable in TensorFlow? -

i make network test save model. code:

import tensorflow tf import numpy np import time  dimensions=100 batch_size=128  def add_layer(inputs, in_size, out_size, activation_function=none):   weights = tf.variable(tf.random_normal([in_size, out_size]))   biases = tf.variable(tf.zeros([1, out_size]) + 0.1)   wx_plus_b = tf.matmul(inputs, weights) + biases   if activation_function none:     outputs = wx_plus_b   else:     outputs = activation_function(wx_plus_b)   return outputs  def f(batch_size,val,dims):   = np.zeros(batch_size,dtype=np.int32)+val   b = np.zeros((batch_size, dims))   b[np.arange(batch_size), a] = 1   return b  xs = tf.placeholder(tf.float32, [none, dimensions]) ys = tf.placeholder(tf.float32, [none, 43])  l1 = add_layer(xs, dimensions, 64, activation_function=none) l2 = add_layer(l1, 64, 64, activation_function=tf.nn.sigmoid) prediction = add_layer(l2, 64, 43, activation_function=none)  loss = tf.reduce_mean(tf.square(ys - prediction)) train_step = tf.train.adamoptimizer(0.003).minimize(loss)  sess = tf.session() sess.run(tf.global_variables_initializer())   step in range(100):   start_time = time.time()   x = f(batch_size=batch_size,val=step,dims=dimensions)   y = np.random.rand(batch_size,43)   sess.run(train_step, feed_dict={xs:x, ys:y})   duration = time.time()-start_time   if step%10 == 0:     loss_value = sess.run(loss, feed_dict={xs: x, ys: y})     format_str = ('step %d,loss=%5.2f (%.1f examples/sec;%.3f sec/batch)')     print(format_str %(step,loss_value,batch_size/duration,float(duration)))        saver = tf.train.saver() save_path = saver.save(sess, "./save_net.ckpt") sess.close()     

it save variables "./save_net.ckpt".

but want save weight , bias of l1 layer. how it?

and how extract these variables in tensorflow?

you should take @ tensorflow documentation. variables

especially part choosing variables save , restore

in case

you should pass name functions creates weights , biases declaration be

weights = tf.variable(tf.random_normal([in_size, out_size]), name=weights_name) biases = tf.variable(tf.zeros([1, out_size]) + 0.1, name = biases_name) 

and then

saver = tf.train.saver({"l1_wieghts": "l1_weights_name",                          "l1_biases": "l1_biases_name",                           "l2_weights":"l2_weights_names",                           "l2_biases":"l2_biases_name"})