Windows 上的 TensorFlow 版本 1.0.0-rc2：“OpKernel（'op：“BestSplits”device_type：“CPU”'）用于未知 op：BestSplits”带有测试代码

问题描述：

我在 Windows 7 SP1 x64 Ultimate（Python 3.5.2 |Anaconda 自定义（64 位））上安装了 TensorFlow 版本 1.0.0-rc2，使用以下命令：

pip install --upgrade https://storage.googleapis.com/tensorflow/windows/cpu/tensorflow-1.0.0rc2-cp35-cp35m-win_amd64.whl

当我尝试在 Eclipse 4.5 或控制台中从https://web.archive.org/web/20170214034751/https://www.tensorflow.org/get_started/os_setup#test_the_tensorflow_installation运行测试脚本时：

import tensorflow as tf
print('TensorFlow version: {0}'.format(tf.__version__))
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
print(sess.run(hello))

我收到一些错误信息：

TensorFlow version: 1.0.0-rc2
'Hello, TensorFlow!'
E c:    f_jenkinshomeworkspace
elease-windevicecpuoswindows    ensorflob
wcorerameworkop_kernel.cc:943] OpKernel ('op: "BestSplits" device_type: "CPU"') for unknown op: BestSplits
E c:    f_jenkinshomeworkspace
elease-windevicecpuoswindows    ensorflowcorerameworkop_kernel.cc:943] OpKernel ('op: "CountExtremelyRandomStats" device_type: "CPU"') for unknown op: CountExtremelyRandomStats
E c:    f_jenkinshomeworkspace
elease-windevicecpuoswindows    ensorflowcorerameworkop_kernel.cc:943] OpKernel ('op: "FinishedNodes" device_type: "CPU"') for unknown op: FinishedNodes
E c:    f_jenkinshomeworkspace
elease-windevicecpuoswindows    ensorflowcorerameworkop_kernel.cc:943] OpKernel ('op: "GrowTree" device_type: "CPU"') for unknown op: GrowTree
E c:    f_jenkinshomeworkspace
elease-windevicecpuoswindows    ensorflowcorerameworkop_kernel.cc:943] OpKernel ('op: "ReinterpretStringToFloat" device_type: "CPU"') for unknown op: ReinterpretStringToFloat
E c:    f_jenkinshomeworkspace
elease-windevicecpuoswindows    ensorflowcorerameworkop_kernel.cc:943] OpKernel ('op: "SampleInputs" device_type: "CPU"') for unknown op: SampleInputs
E c:    f_jenkinshomeworkspace
elease-windevicecpuoswindows    ensorflowcorerameworkop_kernel.cc:943] OpKernel ('op: "ScatterAddNdim" device_type: "CPU"') for unknown op: ScatterAddNdim
E c:    f_jenkinshomeworkspace
elease-windevicecpuoswindows    ensorflowcorerameworkop_kernel.cc:943] OpKernel ('op: "TopNInsert" device_type: "CPU"') for unknown op: TopNInsert
E c:    f_jenkinshomeworkspace
elease-windevicecpuoswindows    ensorflowcorerameworkop_kernel.cc:943] OpKernel ('op: "TopNRemove" device_type: "CPU"') for unknown op: TopNRemove
E c:    f_jenkinshomeworkspace
elease-windevicecpuoswindows    ensorflowcorerameworkop_kernel.cc:943] OpKernel ('op: "TreePredictions" device_type: "CPU"') for unknown op: TreePredictions
E c:    f_jenkinshomeworkspace
elease-windevicecpuoswindows    ensorflowcorerameworkop_kernel.cc:943] OpKernel ('op: "UpdateFertileSlots" device_type: "CPU"') for unknown op: UpdateFertileSlots

为什么？

我在使用 TensorFlow 0.12.1 (安装时) 时没有遇到这样的问题pip install tensorflow==0.12.1：

TensorFlow version: 0.12.1
b'Hello, TensorFlow!'

解决方案 1：

安装今天的夜间版本（CPU 版本）：

pip install --upgrade http://ci.tensorflow.org/view/Nightly/job/nightly-win/85/DEVICE=cpu,OS=windows/artifact/cmake_build/tf_python/dist/tensorflow-1.0.0rc2-cp35-cp35m-win_amd64.whl

已解决问题（不再有 “OpKernel ('op: ”BestSplits“ device_type: ”CPU“') for unknown op: BestSplits”等等）。

目前有一些 SSE 警告：

TensorFlow version: 1.0.0-rc2
b'Hello, TensorFlow!'
2017-02-15 19:56:22.688266: W c:    f_jenkinshomeworkspace
ightly-windevicecpuoswindows    ensorflowcoreplatformcpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE instructions, but these are available on your machine and could speed up CPU computations.
2017-02-15 19:56:22.688266: W c:    f_jenkinshomeworkspace
ightly-windevicecpuoswindows    ensorflowcoreplatformcpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE2 instructions, but these are available on your machine and could speed up CPU computations.
2017-02-15 19:56:22.689266: W c:    f_jenkinshomeworkspace
ightly-windevicecpuoswindows    ensorflowcoreplatformcpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE3 instructions, but these are available on your machine and could speed up CPU computations.
2017-02-15 19:56:22.689266: W c:    f_jenkinshomeworkspace
ightly-windevicecpuoswindows    ensorflowcoreplatformcpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
2017-02-15 19:56:22.689266: W c:    f_jenkinshomeworkspace
ightly-windevicecpuoswindows    ensorflowcoreplatformcpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
2017-02-15 19:56:22.689266: W c:    f_jenkinshomeworkspace
ightly-windevicecpuoswindows    ensorflowcoreplatformcpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.

在这种情况下，您可以尝试如何使用 SSE4.2 和 AVX 指令编译 Tensorflow？

TensorFlow 1.0.0几天前发布了。但是，它有同样的问题。较新的夜间构建有不同的警告：

sess = tf.Session()
2017-02-17 13:01:59.790943: W c:    f_jenkinshomeworkspace
ightly-windevicecpuoswindows    ensorflowcoreplatformcpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE instructions, but these are available on your machine and could speed up CPU computations.

仅供参考： Tensorflow macOS 二进制文件，使用 SSE4.1、SSE4.2 和 AVX 优化进行编译。

要隐藏警告/错误，您可以使用 os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'，例如：

import tensorflow as tf
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
print('TensorFlow version: {0}'.format(tf.__version__))
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
print(sess.run(hello))

TF_CPP_MIN_LOG_LEVEL：

• 0：显示所有日志（这是默认设置）

• 1：过滤掉INFO日志

• 2：另外过滤掉WARNING日志

• 3：另外过滤掉ERROR日志。

解决方案 2：

参考上述建议，我认为采取以下两个步骤很有帮助：

一、升级TensorFlow：

pip install --upgrade tensorflow==1.1.0rc1

然后，错误日志变为警告日志：

W c:    f_jenkinshomeworkspace
elease-windevicecpuoswindows    ensorflowcoreplatformcpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE3 instructions, but these are available on your machine and could speed up CPU computations.

第二，您可能能够抑制 2 级警告过滤器。

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

我认为无需过滤“错误”日志它就能很好地运行。

解决方案 3：

您可能能够抑制级别 2 的警告过滤器。对于我在 virtualenv 安装中安装的 TensorFlow 1.0.1，此方法有效。

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

抱歉，我想补充一下答案，但是我不值得评论。

解决方案 4：

该问题似乎已在 1.1.0rc0 及更高版本中修复。

查找 tensorflow 的最新版本：

pip search --version tensorflow

升级 TensorFlow：

pip install --upgrade tensorflow==1.1.0rc1