Csv 如何在tensorflow 2.0中定义嵌入列？
我是Tensorflow的新手，我使用本地驱动器中的csv数据来学习本教程，我可以加载csv文件并打印列标题Csv 如何在tensorflow 2.0中定义嵌入列？,csv,dataset,tensorflow2.0,Csv,Dataset,Tensorflow2.0,我是Tensorflow的新手，我使用本地驱动器中的csv数据来学习本教程，我可以加载csv文件并打印列标题 for feature_batch, label_batch in train_ds.take(1): print('Every feature:', list(feature_batch.keys())) print('A batch of traffic_type',label_batch) 当我试图用 _mt_datetime_embedding = feature_co
for feature_batch, label_batch in train_ds.take(1):
print('Every feature:', list(feature_batch.keys()))
print('A batch of traffic_type',label_batch)
_mt_datetime_embedding = feature_column.embedding_column(_mt_datetime, dimension=8)
demo(_mt_datetime_embedding)
我不知道怎么了？谁能帮帮我吗？非常感谢。根据Tensorflow关于嵌入列的文档： 假设不是只有几个可能的字符串，而是 每个类别有数千（或更多）个值。由于许多原因，例如 类别的数量越来越多，培训一个 使用一个热编码的神经网络。我们可以使用嵌入列 为了克服这个限制。而不是将数据表示为 一个多维度的热向量，一个嵌入列表示 数据作为低维密集向量，其中每个单元可以 包含任何数字，而不仅仅是0或1 当一个分类列有许多可能的值时，最好使用
嵌入列
tf.feature\u column.embedding\u column
的输入必须是由任何category\u column\u*函数创建的category column
语法：
tf.feature_column.embedding_column(
categorical_column, dimension, combiner='mean', initializer=None,
ckpt_to_load_from=None, tensor_name_in_ckpt=None, max_norm=None, trainable=True,
use_safe_embedding_lookup=True
)
当我将输入添加为数值列
而不是分类列
时，收到属性错误：“数值列”对象没有属性“num\u bucket”
age_embedding = feature_column.embedding_column(age, dimension=8)
demo(age_embedding)
输出：
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
<ipython-input-23-94a5fc74016e> in <module>()
1 age_embedding = feature_column.embedding_column(age, dimension=8)
----> 2 demo(age_embedding)
4 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/python/feature_column/feature_column_v2.py in create_state(self, state_manager)
3181 """Creates the embedding lookup variable."""
3182 default_num_buckets = (self.categorical_column.num_buckets
-> 3183 if self._is_v2_column
3184 else self.categorical_column._num_buckets) # pylint: disable=protected-access
3185 num_buckets = getattr(self.categorical_column, 'num_buckets',
AttributeError: 'NumericColumn' object has no attribute 'num_buckets'
193 train examples
49 validation examples
61 test examples
[[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]
[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]
[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]
[ 0.3212179 0.29932576 -0.44579896 -0.4998746 0.064592 0.16934885
0.02404759 0.5051637 ]
[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]]
输出：
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
<ipython-input-23-94a5fc74016e> in <module>()
1 age_embedding = feature_column.embedding_column(age, dimension=8)
----> 2 demo(age_embedding)
4 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/python/feature_column/feature_column_v2.py in create_state(self, state_manager)
3181 """Creates the embedding lookup variable."""
3182 default_num_buckets = (self.categorical_column.num_buckets
-> 3183 if self._is_v2_column
3184 else self.categorical_column._num_buckets) # pylint: disable=protected-access
3185 num_buckets = getattr(self.categorical_column, 'num_buckets',
AttributeError: 'NumericColumn' object has no attribute 'num_buckets'
193 train examples
49 validation examples
61 test examples
[[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]
[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]
[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]
[ 0.3212179 0.29932576 -0.44579896 -0.4998746 0.064592 0.16934885
0.02404759 0.5051637 ]
[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]]
有关更多详细信息，请参考Tensorflow文档中关于嵌入列的部分：
假设不是只有几个可能的字符串，而是
每个类别有数千（或更多）个值。由于许多原因，例如
类别的数量越来越多，培训一个
使用一个热编码的神经网络。我们可以使用嵌入列
为了克服这个限制。而不是将数据表示为
一个多维度的热向量，一个嵌入列表示
数据作为低维密集向量，其中每个单元可以
包含任何数字，而不仅仅是0或1
当一个分类列有许多可能的值时，最好使用嵌入列
tf.feature\u column.embedding\u column
的输入必须是由任何category\u column\u*函数创建的category column
语法：
tf.feature_column.embedding_column(
categorical_column, dimension, combiner='mean', initializer=None,
ckpt_to_load_from=None, tensor_name_in_ckpt=None, max_norm=None, trainable=True,
use_safe_embedding_lookup=True
)
当我将输入添加为数值列
而不是分类列
时，收到属性错误：“数值列”对象没有属性“num\u bucket”
age_embedding = feature_column.embedding_column(age, dimension=8)
demo(age_embedding)
输出：
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
<ipython-input-23-94a5fc74016e> in <module>()
1 age_embedding = feature_column.embedding_column(age, dimension=8)
----> 2 demo(age_embedding)
4 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/python/feature_column/feature_column_v2.py in create_state(self, state_manager)
3181 """Creates the embedding lookup variable."""
3182 default_num_buckets = (self.categorical_column.num_buckets
-> 3183 if self._is_v2_column
3184 else self.categorical_column._num_buckets) # pylint: disable=protected-access
3185 num_buckets = getattr(self.categorical_column, 'num_buckets',
AttributeError: 'NumericColumn' object has no attribute 'num_buckets'
193 train examples
49 validation examples
61 test examples
[[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]
[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]
[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]
[ 0.3212179 0.29932576 -0.44579896 -0.4998746 0.064592 0.16934885
0.02404759 0.5051637 ]
[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]]
输出：
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
<ipython-input-23-94a5fc74016e> in <module>()
1 age_embedding = feature_column.embedding_column(age, dimension=8)
----> 2 demo(age_embedding)
4 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/python/feature_column/feature_column_v2.py in create_state(self, state_manager)
3181 """Creates the embedding lookup variable."""
3182 default_num_buckets = (self.categorical_column.num_buckets
-> 3183 if self._is_v2_column
3184 else self.categorical_column._num_buckets) # pylint: disable=protected-access
3185 num_buckets = getattr(self.categorical_column, 'num_buckets',
AttributeError: 'NumericColumn' object has no attribute 'num_buckets'
193 train examples
49 validation examples
61 test examples
[[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]
[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]
[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]
[ 0.3212179 0.29932576 -0.44579896 -0.4998746 0.064592 0.16934885
0.02404759 0.5051637 ]
[-0.4675103 0.61985296 0.06297898 0.00818724 0.05449321 -0.6865342
-0.05250816 -0.13339798]]
有关更多详细信息，请参阅