我是PyTorch的新手。我想跟踪我的模型通过其优化的参数空间中的距离。这是我正在使用的代码。
class ParameterDiffer(object):
def __init__(self, network):
network_params = []
for p in network.parameters():
network_params.append(p.data.numpy())
self.network_params = network_params
def get_difference(self, network):
total_diff = 0.0
for i, p in enumerate(network.parameters()):
p_np = p.data.numpy()
diff = self.network_params[i] - p_np
# print(diff)
scalar_diff = np.sum(diff ** 2)
total_diff += scalar_diff
return total_diff
这会有用吗？我跟踪total_diff的时间，并记录它，但似乎总是零。尽管该模型的性能正在提高，但这让我感到非常困惑。
答案 0 :(得分：1)
这是因为PyTorch处理numpy数组和火炬'''
Developer: Roei Edri
File name: trcrt.py
Date: 24.11.17
Version: 1.1.0
Description: Get an url as an input and prints the traceroute to it.
'''
import sys
import urllib2
i, o, e = sys.stdin, sys.stdout, sys.stderr
from scapy.all import *
from scapy.layers.inet import *
sys.stdin, sys.stdout, sys.stderr = i, o, e
def trcrt(dst):
"""
Check for the route for the given destination
:param dst: Final destination, in a form of a website.
:type dst: str
"""
try:
pckt = IP(dst=dst)/ICMP() # Creates the
# packet
ip = [p for p in pckt.dst] # Gets the ip
print "Tracerouting for {0} : {1}".format(dst, ip[0])
for ttl in range(1, 40):
pckt = IP(ttl=ttl, dst=dst)/ICMP()
timeBefore = time.time()
reply = sr1(pckt, verbose=0, timeout=5)
timeAfter = time.time()
timeForReply = (timeAfter - timeBefore)*1000
if reply is not None:
print "{0} : {1} ; Time for reply: {2}".format(ttl,
reply.src, timeForReply)
if reply.type == 0:
print "Tracerout Completed"
break
else:
print "{0} ... Request Time Out".format(ttl)
def checkInternet():
"""
Checks if there is an internet connection
:return: True if there is an internet connection
"""
try:
urllib2.urlopen('http://45.33.21.159', timeout=1)
return True
except urllib2.URLError as IntError:
return False
之间的转换的方式。如果numpy数组和torch Tensor之间的基础数据类型相同，它们将共享内存。更改一个值也会改变另一个的值。我将在这里展示一个具体的例子，
Tensor然后，如果您更改x 就地并查看x和y中的值，您会发现它们仍然相同。
x = Variable(torch.rand(2, 2))
y = x.data.numpy()
x
Out[39]:
Variable containing:
0.8442 0.9968
0.7366 0.4701
[torch.FloatTensor of size 2x2]
y
Out[40]:
array([[ 0.84422851, 0.996831 ],
[ 0.73656738, 0.47010136]], dtype=float32)
因此，在模型更新期间，模型和类x += 2
x
Out[42]:
Variable containing:
2.8442 2.9968
2.7366 2.4701
[torch.FloatTensor of size 2x2]
y
Out[43]:
array([[ 2.84422851, 2.99683094],
[ 2.7365675 , 2.47010136]], dtype=float32)
中的参数将始终相同。这就是你看到零的原因。
如果numpy数组和火炬Tensor的基础数据类型不兼容，它将强制在火炬Tensor中复制原始数据，这将使numpy数组和火炬Tensor具有单独的内存。
一种简单的方法就是将numpy数组转换为ParameterDiffer类型。而不是
np.float64您可以使用
network_params.append(p.data.numpy())