Python3.6 代码运行错误 about batchnorm_prune HOT 7 CLOSED

can you try python2.7?

I developed this code using python2.7 + tensorflow1.14. I was able to follow the README with no crashes.

Its behavior under python3.6 is unknown to me. I know the python2 will soon be deprecated. I will try to fix the issue for python3.5 or higher.

from batchnorm_prune.

python2.7的环境我还没尝试，python3.6是可以训练，就是到最后一次，训练完成开始剪枝的时候报错在collection.append(tf.py_func(lambda a, b: self.flow.get_channel_pruned(a, b),[tf.constant(varname), v], tf.bool)) 这儿，我很困惑。

from batchnorm_prune.

It may be the pickle library is incompatible between python 2 and 3. Will let you know if I figured out.

from batchnorm_prune.

谢谢你耐心指导，pickle我已经在导包时候，换成python3.x的写法，你可以试试，在python3.x环境中，是不是与我同样的问题，方便话传一份python3.x环境代码，再次感谢。

from batchnorm_prune.

Hi,我又调试了一遍在Graph.py文件中get_channel_pruned添加了 varname = str(varname, encoding="utf8")，因为在我python3.6环境中出现varname为二进制，
KeyError: b'g1b1c0/BatchNorm/gamma'这样的错误，经过转换后，现在代码可以完全运行，没有任何错误，不知道我改的对不对，谢谢。
代码如下：

def get_channel_pruned(self, varname, v):
varname = str(varname, encoding="utf8")
if len(self.edgename_map[varname]) == 2:
v_in, v_out = self.edgename_map[varname]
for i in range(v.shape[0]):
if v[i]:
v1 = v_in + "%d" % i
v2 = v_out + "%d" % i
self.nodes[v1].o.remove(v2)
self.nodes[v2].i.remove(v1)

from batchnorm_prune.

from batchnorm_prune.

@zaf05 Hi, could you verify if you can run the latest commit?

from batchnorm_prune.