如何在Kaggle笔记本上安装“树集成层”

Question

我想在Kaggle笔记本上尝试以下代码，但我找不到安装tf_树的方法

from tensorflow import keras
from tf_trees import TEL

tree_layer = TEL(output_logits_dim=2, trees_num=10, depth=3)

model = keras.Sequential()
model.add(keras.layers.BatchNormalization())
model.add(tree_layer)

似乎tf_树无法使用安装！pip安装

如果有人能提出解决方案，我将不胜感激。多谢各位

排序：https://github.com/google-research/google-research/tree/master/tf_trees

Answer 1

首先打开internet支持并从github克隆google research repo：

!git clone https://github.com/google-research/google-research.git

然后我们需要g++的编译和链接选项，因此请运行以下代码段：

import tensorflow as tf; 
print(" ".join(tf.sysconfig.get_compile_flags()))

及

import tensorflow as tf; 
print(" ".join(tf.sysconfig.get_link_flags()))

对于我的笔记本，我得到了以下标志：

-I/opt/conda/lib/python3.7/site-packages/tensorflow/include -D_GLIBCXX_USE_CXX11_ABI=0
-L/opt/conda/lib/python3.7/site-packages/tensorflow -l:libtensorflow_framework.so.2

然后，用上述输出替换变量${TF_CFLAGS[@]}和${TF_LFLAGS[@]}

!g++ -std=c++11 -shared google-research/tf_trees/neural_trees_ops.cc google-research/tf_trees/neural_trees_kernels.cc google-research/tf_trees/neural_trees_helpers.cc -o google-research/tf_trees/neural_trees_ops.so -fPIC -I/opt/conda/lib/python3.7/site-packages/tensorflow/include -D_GLIBCXX_USE_CXX11_ABI=0 -L/opt/conda/lib/python3.7/site-packages/tensorflow -l:libtensorflow_framework.so.2 -O2

最后我们需要添加系统路径

import sys
sys.path.insert(1, '/kaggle/working/google-research')

然后运行您的代码片段

from tensorflow import keras
from tf_trees import TEL

tree_layer = TEL(output_logits_dim=2, trees_num=10, depth=3)

model = keras.Sequential()
model.add(keras.layers.BatchNormalization())
model.add(tree_layer)