Skip to main content

Share a session

A whole model session can leave dashAI as a single .dashai file and be imported in another dashAI, or in the same one. Use it to hand out a session to a class, pass your work to a teammate, move sessions to another computer, or let someone retrain your experiment and compare.

Export​

In the left panel, open the menu (...) of a session and choose Export. The dialog lets you:

  • Choose the runs to include. All of them are selected; runs still training cannot be exported until they finish.
  • Include trained models, checked by default. Uncheck it to export only the configuration, without trained models: the dataset and the settings of the session and its runs. The file is smaller, and whoever imports it trains the models again.

Finished runs travel with their model and metrics. Runs that failed or were never trained travel without a model and arrive untrained.

Import​

Next to the Sessions title in the left panel, click the import button, read the warning and choose the .dashai file. dashAI creates a new session with its own copy of the dataset and opens it. From there you can predict, retrain and add models as in any other session.

Importing never changes what you already have. If a session or dataset with the same name exists, the new one gets a number, such as iris (2).

What the file contains​

  • The dataset, including its train, validation and test split.
  • The session configuration: task, columns, preprocessing and evaluation settings.
  • Each chosen run's configuration and, when included, its trained model, metrics and the fitted preprocessing.

Explanations, predictions and reports are not included yet; generate them again after importing.

Compatibility​

  • A file from a newer dashAI is refused: update dashAI to open it.
  • If the session uses a plugin you do not have, the import is refused and the message says which one to install.
  • If your library versions differ from the ones used to export, the session is imported with a warning. If a model then fails, use Retrain: the configuration is always in the file.
  • If a model must be downloaded (for example a HuggingFace translator) and you do not have it, dashAI tells you after importing. A trained run still predicts from its own copy, and you only need the download to retrain it. A run that arrives untrained needs it before training, and the notice offers a Download button.
warning

A session file contains Python objects, and using its models runs code. Import only sessions from people you trust.

note

The file always includes the dataset, and some models (for example KNN) also keep training rows inside them. Do not share a session trained on private data.