How to Add Your Own Modeltype

Modeltype Category

TrainDB uses two modeltypes for approximate query processing: synopsis generation modeltype and inference modeltype. The synopsis generation modeltype generates synthetic data similar to the original data, and the inference modeltype predicts approximate results of aggregate functions. You can add your own modeltype by referring to the following section, depending on the category of your modeltype.

Implementing a Modeltype

You can add your own modeltype by inheriting and implementing the abstract base class. You have to implement the following classes depending on the cateogy of your modeltype.

class TrainDBSynopsisModel(TrainDBModel, abc.ABC):
  """Base class for all the ``TrainDB`` synopsis generation models."""

  def synopsis(self, row_count):
    pass

class TrainDBInferenceModel(TrainDBModel, abc.ABC):

  """Base class for all the ``TrainDB`` inference models."""

  def infer(self, agg_expr, group_by_column, where_condition):
    pass

Since the above two modeltypes inherit a common base class, the functions included the base class must also be implemented.

class TrainDBModel(abc.ABC):
  """Base class for all the ``TrainDB`` models."""

  ...

  def train(self, real_data, table_metadata):
    pass


  def save(self, output_path):
    pass


  def load(self, input_path):
    pass


  def list_hyperparameters():
    return []