我试图在TensorFlow卷积网络中可视化激活。但是，我似乎对相同的输入数据进行了不同的激活。如果我有一些功能和一个函数<input type="radio" name="cash" id="cash" value="CASH"/>CASH<br />
<input type="radio" name="cash" id="card" value="CARD"/>CARD<br />
<input type="radio" name="cash" id="netbank" value="NETBANKING"/>NETBANKING<br />
<input type="radio" name="cash" id="paypal" value="PAYPAL"/>PAYPAL<br />
，它会创建一个输入张量并运行以下两次：
get_input_tensors我为第一次运行得到了这个输出：
data = get_input_tensors(features)
convolved = tf.layers.conv1d(
data,
filters=params.num_conv[i],
kernel_size=params.conv_len[i],
activation=None,
strides=1,
padding="same",
name="conv1d_%d" % i)
with tf.Session() as sess:
activations = sess.run(data)
print(activations)
with tf.Session() as sess:
init_g = tf.global_variables_initializer()
sess.run(init_g)
activations = sess.run(convolved)
print(activations)
第二次运行：
[[[ 0.33333334 1. 0. ]
[-0.33333334 -0.37117904 0. ]
[ 1. -0.62882096 1. ]
[ 0. 0. 0. ]
[ 0. 0. 0. ]
[ 0. 0. 0. ]
[ 0. 0. 0. ]
[ 0. 0. 0. ]
[ 0. 0. 0. ]
[ 0. 0. 0. ]]
[[ 0. 1. 0. ]
[ 0.37542662 -0.02620087 0. ]
[ 0.09215017 -0.09170306 0. ]
[-0.41638225 -0.3580786 0. ]
[-0.00341297 -0.01310044 0. ]
[ 0.82935154 -0.17467248 0. ]
[ 0.12286689 -0.17030568 0. ]
[-0.23890784 -0.15283842 0. ]
[-0.46075085 -0.01310044 0. ]
[-0.04095563 0.01746725 1. ]]]
[[[-0.8071091 -0.23191781 0.13636628 -0.69688106]
[ 0.39058334 -0.14330778 0.4304243 0.25608253]
[ 0.14675646 -0.520292 0.34630966 1.2224951 ]
[ 0.759295 0.8370328 -0.13724771 0.22211897]
[ 0. 0. 0. 0. ]
[ 0. 0. 0. 0. ]
[ 0. 0. 0. 0. ]
[ 0. 0. 0. 0. ]
[ 0. 0. 0. 0. ]
[ 0. 0. 0. 0. ]]
[[-0.35912833 0.31992826 -0.27506 -0.42530814]
[-0.19117472 -0.18537153 0.5497088 -0.23093367]
[-0.07501456 -0.2450811 0.35258675 -0.2551663 ]
[ 0.37794912 0.06009946 -0.03035221 0.09803987]
[-0.10526019 0.26594305 -0.43844843 0.33906972]
[-0.36485478 -0.16686419 0.18421796 0.24412222]
[ 0.28276905 0.08124011 0.24421532 -0.09371081]
[ 0.13729642 -0.1578648 0.07745218 -0.07478261]
[ 0.13861918 -0.41384116 -0.2183905 0.49029657]
[ 0.29436743 -0.3423192 0.2173931 0.55723166]]]
为什么这些不一样？
编辑：我已将第二个[[[ 0.33333334 1. 0. ]
[-0.33333334 -0.37117904 0. ]
[ 1. -0.62882096 1. ]
[ 0. 0. 0. ]
[ 0. 0. 0. ]
[ 0. 0. 0. ]
[ 0. 0. 0. ]
[ 0. 0. 0. ]
[ 0. 0. 0. ]
[ 0. 0. 0. ]]
[[ 0. 1. 0. ]
[ 0.37542662 -0.02620087 0. ]
[ 0.09215017 -0.09170306 0. ]
[-0.41638225 -0.3580786 0. ]
[-0.00341297 -0.01310044 0. ]
[ 0.82935154 -0.17467248 0. ]
[ 0.12286689 -0.17030568 0. ]
[-0.23890784 -0.15283842 0. ]
[-0.46075085 -0.01310044 0. ]
[-0.04095563 0.01746725 1. ]]]
[[[-0.28877693 0.47691846 -0.08552396 -0.06404732]
[-0.6436144 -0.88326526 0.07262921 0.7572223 ]
[-0.38767207 -0.68364584 -0.5907324 -0.84791625]
[-0.77412176 0.17703977 0.19723669 -0.1314309 ]
[ 0. 0. 0. 0. ]
[ 0. 0. 0. 0. ]
[ 0. 0. 0. 0. ]
[ 0. 0. 0. 0. ]
[ 0. 0. 0. 0. ]
[ 0. 0. 0. 0. ]]
[[-0.06407069 0.46260497 0.15592983 0.01113943]
[-0.17851508 -0.21558103 -0.12748975 0.15542175]
[-0.34141034 -0.05112889 -0.18030314 -0.01680391]
[ 0.15434177 -0.08103228 -0.04429122 0.12980668]
[-0.00583593 -0.01706403 -0.02277096 -0.15864758]
[-0.51721746 -0.30063802 -0.06769364 -0.36139038]
[-0.53160226 -0.16488285 0.0127665 -0.1110348 ]
[ 0.10733443 0.04029365 -0.04993725 0.07187385]
[ 0.0712772 -0.37336436 0.36313307 0.5290657 ]
[ 0.23458107 -0.12172135 -0.59520864 -0.27075604]]]
更改为：
with tf.Session()...但是我收到以下错误：
with tf.Session() as sess:
saver = tf.train.import_meta_graph('models/conv_model/model.ckpt-100000.meta')
saver.restore(sess, tf.train.latest_checkpoint('models/conv_model/'))
outputTensors = sess.run(convolved)
答案 0 :(得分：0)
tf.global_variables_initializer()将初始化所有变量。这意味着它将运行随机数生成器以根据其初始化器生成卷积权重。在你的情况下，你运行它两次，所以你会得到不同的权重随机数。
如果您希望它具有可重现性，则只需运行初始化程序一次，保存会话（包含变量）并修改测试以加载已保存的会话。