关于vocabulary feature 转numerical feature的疑惑

abbiew · June 24, 2021, 3:17am

想请教一下在这个文件第 281 行具体是怎么进行vocabulary feature对numerical feature的转化的：google-research/adversarial_reweighting_model.py at d154ffb7a92f7899807496cb8cb665fb5a01e65c · abbie0214/google-research · GitHub

从281行我一直定位到了以下文件的2461行：

github.com

tensorflow/tensorflow/blob/a4dfb8d1a71385bd6d122e4f27f86dcebb96712d/tensorflow/python/feature_column/feature_column.py#L2461


      
          source_dimension = self.source_column.shape[0]
          
          
i1 = array_ops.reshape(
              array_ops.tile(
                  array_ops.expand_dims(math_ops.range(0, batch_size), 1),
                  [1, source_dimension]),
              (-1,))
          i2 = array_ops.tile(math_ops.range(0, source_dimension), [batch_size])
          # Flatten the bucket indices and unique them across dimensions
          # E.g. 2nd dimension indices will range from k to 2*k-1 with k buckets
          bucket_indices = (
              array_ops.reshape(input_tensor, (-1,)) +
              (len(self.boundaries) + 1) * i2)
          
          
indices = math_ops.cast(
              array_ops.transpose(array_ops.stack((i1, i2))), dtypes.int64)
          dense_shape = math_ops.cast(
              array_ops.stack([batch_size, source_dimension]), dtypes.int64)
          sparse_tensor = sparse_tensor_lib.SparseTensor(
              indices=indices,
              values=bucket_indices,

可是觉得这一行看起来也不像是在从vocabulary feature转成numerical feature，我尝试tf.session()进行打印，但是都跳过了这一行，不知道问题在哪里。想请教一下！谢谢！

snowkylin · June 24, 2021, 7:27am

我没有用过Feature Columns，不过可以参考一下 tf.feature_column的特征处理探究 - 知乎