特征列可以视为原始数据与Estimator之间的桥梁。比如，在Iris的DNNClassifier分类的示例中，特征列通过tf.feature_column.numeric_column创建。DNN等Estimator需要使用数值进行运算，但现实中的特征列并不完全是数值类型。

要创建特征列，需要使用tf.feature_column

Numberic column

在Iris分类中即是调用numeric_column

# Defaults to a tf.float32 scalar.
numeric_feature_column = tf.feature_column.numeric_column(key="SepalLength")

设置dtype

# Represent a tf.float64 scalar.
numeric_feature_column = tf.feature_column.numeric_column(key="SepalLength",
                                                          dtype=tf.float64)

修改shape

# Represent a 10-element vector in which each cell contains a tf.float32.
vector_feature_column = tf.feature_column.numeric_column(key="Bowling",
                                                         shape=10)

# Represent a 10x5 matrix in which each cell contains a tf.float32.
matrix_feature_column = tf.feature_column.numeric_column(key="MyMatrix",
                                                         shape=[10,5])

Bucketized column

有时，你可能不想将值直接用于model，而是将其分为几类。比如，将房子的构建日期分类

而bucket的表示如下：

上面将一个单独的数值，分为4个元素的向量。此时，模型可以通过4个独立的权重学习，而不再仅仅是一个。

# First, convert the raw input to a numeric column.
numeric_feature_column = tf.feature_column.numeric_column("Year")

# Then, bucketize the numeric column on the years 1960, 1980, and 2000.
bucketized_feature_column = tf.feature_column.bucketized_column(
    source_column = numeric_feature_column,
    boundaries = [1960, 1980, 2000])

Categorical identity column

Categorical identity columns可以视为特殊的bucketized columns。

注意，这是一个one-hot encoding，而不是binary numerical encoding。

# Create categorical output for an integer feature named "my_feature_b",
# The values of my_feature_b must be >= 0 and < num_buckets
identity_feature_column = tf.feature_column.categorical_column_with_identity(
    key='my_feature_b',
    num_buckets=4) # Values [0, 4)

# In order for the preceding call to work, the input_fn() must return
# a dictionary containing 'my_feature_b' as a key. Furthermore, the values
# assigned to 'my_feature_b' must belong to the set [0, 4).
def input_fn():
    ...
    return ({ 'my_feature_a':[7, 9, 5, 2], 'my_feature_b':[3, 1, 2, 2] },
            [Label_values])

Categorical vocabulary column

对于strings，我们必须将其转换为numeric或categorical values

vocabulary_feature_column =
    tf.feature_column.categorical_column_with_vocabulary_list(
        key="a feature returned by input_fn()",
        vocabulary_list=["kitchenware", "electronics", "sports"])

当字典值很多时，可以使用

vocabulary_feature_column =
    tf.feature_column.categorical_column_with_vocabulary_file(
        key="a feature returned by input_fn()",
        vocabulary_file="product_class.txt",
        vocabulary_size=3)

而product_class.txt内容为

kitchenware
electronics
sports

Hashed Columns

hashed_feature_column =
    tf.feature_column.categorical_column_with_hash_bucket(
        key = "some_feature",
        hash_buckets_size = 100)

Crossed column

将多个特征合并为单个特征，从而让model可以对不同feature集合进行学习。

我们知道，城市中土地价格与其位置相关，但表示位置的经度和维度，当单独与价格关联时，并没有用处。

def make_dataset(latitude, longitude, labels):
    assert latitude.shape == longitude.shape == labels.shape

    features = {'latitude': latitude.flatten(),
                'longitude': longitude.flatten()}
    labels=labels.flatten()

    return tf.data.Dataset.from_tensor_slices((features, labels))

# Bucketize the latitude and longitude usig the `edges`
latitude_bucket_fc = tf.feature_column.bucketized_column(
    tf.feature_column.numeric_column('latitude'),
    list(atlanta.latitude.edges))

longitude_bucket_fc = tf.feature_column.bucketized_column(
    tf.feature_column.numeric_column('longitude'),
    list(atlanta.longitude.edges))

# Cross the bucketized columns, using 5000 hash bins.
crossed_lat_lon_fc = tf.feature_column.crossed_column(
    [latitude_bucket_fc, longitude_bucket_fc], 5000)

fc = [
    latitude_bucket_fc,
    longitude_bucket_fc,
    crossed_lat_lon_fc]

# Build and train the Estimator.
est = tf.estimator.LinearRegressor(fc, ...)

Indicator和embedding columns

Indicator columns和embedding columns通常会将categorical columns作为输入

特征列