jupyter笔记本固定在使用matplotlib（CPU 100％）绘制张量（Tensorflow 2）上
我正在关注视频教程，并使用matplotlib在jupyter笔记本中绘制张量张量。 牢房刚刚卡住，一个CPU获得了100％的速度。
import tensorflow as tf
import tensorflow_probability as tfp
tfd = tfp.distributions
tfb = tfp.bijectors
import matplotlib.pyplot as plt
# %matplotlib inline
normal = tfd.Normal(loc=0, scale=1)
n = 10000
z = normal.sample(n)
scale = 4.5
shift = 7
scale_and_shift = tfb.Chain([tfb.Shift(shift), tfb.Scale(scale)])
x = scale_and_shift.forward(z)
plt.hist(z, bins=60, density=True)
plt.show()
在我关注的教程中，它运行顺利。但是在我的尝试中，它被卡住了。 为什么在我的情况下它不起作用？我应该安装任何软件包吗？ z是张量，可以直接绘制它吗？
我注意到如果我使用plt.hist.hist（z.numpy（），bins = 60，密度= true）它起作用。但是仍然想知道为什么绘制z直接在我的环境中不起作用。
I'm following a video tutorial and using matplotlib to plot a tensorflow tensor in a Jupyter Notebook.
The cell just stuck and one CPU went 100%.
import tensorflow as tf
import tensorflow_probability as tfp
tfd = tfp.distributions
tfb = tfp.bijectors
import matplotlib.pyplot as plt
# %matplotlib inline
normal = tfd.Normal(loc=0, scale=1)
n = 10000
z = normal.sample(n)
scale = 4.5
shift = 7
scale_and_shift = tfb.Chain([tfb.Shift(shift), tfb.Scale(scale)])
x = scale_and_shift.forward(z)
plt.hist(z, bins=60, density=True)
plt.show()
In the tutorial I'm following, it runs smoothly. But in my attempt it got stuck. Why it doesn't work in my case? Is there any package I should install? z is a tensor, is it OK to plot it directly?
I noticed if I use plt.hist(z.numpy(), bins=60, density=True) it works. But still wondering why plotting z directly does not work in my environment.
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（1）
首先，这不是丢失的软件包问题，通常直接绘制张量没有问题。例如，您可以完成
plt.plot（z）而没有问题。如果您检查plt.hist.hista>您可以看到这个：在您的情况下，由于您正在绘制分布中的样本值（
z是[10000]形状张量），因此您在第二类中（[N0，N1，...]）实际上，您正在创建10000个重叠直方图。为了清楚起见，请参见s = 4：这就是为什么您的CPU被超载 - 尝试一次创建10000个直方图（也就是为什么每行都是不同的颜色 - ＆gt;不同的直方图）。
当您使用
Tensor.numpy（）时，它将张量转换为使用Numpy数组，并能够正确创建直方图。PS不确定为什么它在教程中工作。也许Trey're使用不同的软件包的版本 - 但是
z.numpy（）应该给您相同的结果Firstly, that's not a missing package problem and generally there's no problem in plotting tensors directly. For example you could have done
plt.plot(z)without as issue. If you checkplt.histdocumentation you can see this:In your case, since you're plotting sample values from the distribution (with
zbeing a [10000] shaped tensor), you are on the second category ([n0, n1, ...]) and you're actually creating 10000 overlapping histograms. To make it clear, see this fors=4:That's why your CPU is overloaded - trying to create 10000 histograms at once (also thats why each line is a different color --> different histograms).
When you use
tensor.numpy()it converts the Tensor to a NumPy array using and is able to create the histogram correctly.PS not sure why it's working in the tutorial. Maybe trey're using different packages' version - but
z.numpy()should give you the same results