FAQ

DeepFlow is a modular software platform. Users can use it with multiple modules according to different development process stages, and it has rich and complete functions.

• Data management and access:

  It provides the function of project and data isolation. Workspaces can be set up on the platform according to the project, and each workspace is equipped with an independent data storage space. It also provides a “Data Profiling” function to help data scientists understand data dependencies.

• Model development:

  Through the “Containers”, data scientists can choose their favorite IDE environment (for example: Jupyter, VScode, Rstudio) to start developing models immediately. Data scientists who are not programmers can also use the built-in “AutoML” function. No code / low code completes model development.

• Model training:

  The system’s “Hyperparameter Tuning” module can parallelize existing models to perform multiple parameter training and provide suggestions for the best parameter combinations; in addition, real-time or regular targeting can be set through the “Job Scheduler” The model is retrained to ensure that the model continues to be healthy.

• Model management and deployment:

  The “Model Repository” module is like the production history of the model. Information such as the input and output and characteristic values ​​of each model version can be accessed at any time. Data scientists can select appropriate models from the “Model Repository” for model inference; in the “Model Deploy” module, model inference can be quickly completed by turning on and off with one click. In addition, multiple different versions of the same model can be deployed at the same time for easy validity comparison.