Frequently asked questions

Does my data leave the instance?

No. Files you upload are held in memory on your instance, not saved to its disk, and never sent anywhere. The Studio has no accounts, no usage tracking and no licence check, and makes no calls to Software Sushi. See Security and data.

Is my work saved?

No. The Studio does not keep projects between sessions. Download the model, the scored file and the training code before you close the tab. To keep every model you trained, not only the one you picked, use Download every model on Predict & export.

What kinds of question can it answer?

Two kinds, on tabular data: sorting rows into groups, such as which customers will cancel, and predicting a number, such as a sale price. It does not do time-series forecasting, or work with images or free text.

How large a file can it handle?

Up to 50 MB and 500 columns. When training a large file will take a while, the Studio estimates how long before you start.

Can I use the model outside the Studio?

Yes. The download is a joblib file holding a standard scikit-learn pipeline. It loads in any Python environment with scikit-learn, pandas and joblib, with no dependency on Sushi ML Studio. See Score a file in Python.

Can several people use it at once?

Yes, with the one password. Each browser tab is its own session, so people do not see each other's work. They share the instance's processors, so training is slower when several people train at the same time.

Can I use HTTPS?

Not directly. Reach the Studio through an SSH or Session Manager tunnel, which is encrypted, as described in Reach the Studio privately. Or place it behind a load balancer or reverse proxy that holds your certificate.

How do I move to a newer version?

Launch a new instance from the newer version on AWS Marketplace. The Studio keeps nothing on the instance apart from its password, so there is nothing to move. Copy across any notebooks or files you saved in JupyterLab, then terminate the old instance.

How do I stop paying for it?

Stop the instance when you are not using it, and start it again later. A stopped instance is not charged for compute or software, only for its disk, and for an Elastic IP if you attached one. When you are finished for good, terminate it, then cancel the subscription in AWS Marketplace.

Does it need internet access?

Not for the Studio itself. Installing extra Python packages with pip does.