Tensorflow：如何为 numpy 矩阵输入创建 feature_columns

Question

我正在使用 tensorflow 1.8.0、python 3.6.5。 数据是鸢尾花数据集。代码如下：

import pandas as pd
import numpy as np
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
import tensorflow as tf

X = iris['data']
y = iris['target']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

input_train=tf.estimator.inputs.numpy_input_fn(x=X_train,
            y=y_train, num_epochs=100, shuffle=False)
classifier_model = tf.estimator.DNNClassifier(hidden_units=[10, 
                  20, 10], n_classes=3, feature_columns=??)

这是我的问题，如何为 numpy 矩阵设置 feature_columns？

如果我将 X 和 y 转换为 pandas.DataFrame，我可以对 feature_columns 使用以下代码，它适用于 DNNClassifier 模型。

features = X.columns
feature_columns = [tf.feature_column.numeric_column(key=key) for key in features]

Answer 1

您可以将您的 numpy ndarray 包装在字典中并将其作为输入 x 传递给 numpy_input_fn 方法，然后使用该字典中的键来定义您的 feature_column。还要注意，因为你的X_train中的每一个数据都有4个维度，所以在定义tf.feature_column.numeric_column的时候需要指定shape参数。这是完整的代码：

import pandas as pd
import numpy as np
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
import tensorflow as tf

iris = load_iris()

X = iris['data']
y = iris['target']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

input_train = tf.estimator.inputs.numpy_input_fn(
                  x = {'x': X_train},
                  y = y_train,
                  num_epochs = 100,
                  shuffle = False)

feature_columns = [tf.feature_column.numeric_column(key='x', shape=(X_train.shape[1],))]

classifier_model = tf.estimator.DNNClassifier(
                       hidden_units=[10, 20, 10],
                       n_classes=3,
                       feature_columns=feature_columns)