MLModel_en.md

MLModel

---

MLModel is the complete model of a machine-learning algorithm. It serves as the base class for all machine learning models on Angel. A complex machine-learning algorithm usually needs multiple PSModel to work together, and MLModel provides unified handling of these PSModels.

Functionality

MLModel is an abstract class and needs to be inherited in implementing a specific machine-learning algorithm. It is a container class that handles all the PSModels in the algorithm; all the PSModels in it will be loaded, trained and saved as an overall model. In addition, MLModel contains the logic of the predict method.

Core Methods

• predict

  • Definition: def predict(storage: DataBlock[predictType]): DataBlock[PredictResult]
  • Functionality: get model predictions (given a trained model); procedures of this method need to be implemented in specific PSModel
  • Parameters: storage: DataBlock[predictType], data
  • Return value: DataBlock[PredictResult], prediction result on provided data

• setSavePath

  • Definition: def setSavePath(conf: Configuration)
  • Functionality: set save path for model
  • Parameters: conf: Configuration, job configuration
  • Return value: none

• setLoadPath

  • Definition: def setLoadPath(conf: Configuration)
  • Functionality: set load path for model
  • Parameters: conf: Configuration, job configuration
  • Return value: none

• addPSModel

  • Definition: def addPSModel(name: String, psModel: PSModel[_])
  • Functionality: add a new PSModel to MLModel
  • Parameters:
    • name: String, PSModel name
    • psModel: PSModel[_], PSModel object
  • Return value: none

• getPSModel

  • Definition: def getPSModel(name: String): PSModel[_]
  • Functionality: retrieve PSModel object with its name
  • Parameters: name: String, PSModel name
  • Return value: a PSModel object