Experiment execution
What is an experiment?
An experiment is a script run on one or more datasets. It produces result files, or an error if something goes wrong. To run an experiment, open an existing project (Projects) or create a new project and choose Run Experiment.
Create and run an experiment
Select a script and one or more datasets from the lists.
You can choose scripts or datasets that:
- You own, or
- You have been granted access to, or
- Other project collaborators have already added to the project.
A teal icon means the script or dataset is already in the project. A rose icon means you can access it, but it is not in the project yet. Selecting a rose item adds it to the project automatically.
After you select the script and dataset(s), review the estimated cost in credits. You can use your credits or, if your organization allows it, your organization’s credits. Then choose Run.
If the script uses more than one dataset, put them in the order your script expects. That order is what RAISE_DATASET_ID_LIST uses. See Dynamic dataset accessing.
Statuses
While the experiment runs you can watch its status, or open error details if it fails.
| Status | Meaning |
|---|---|
| Awaiting Execution | Waiting on the node |
| Transferring Datasets | Datasets are being copied to the node if needed |
| Dataset Transfer Failed | That copy failed |
| Script Analysis Pending / Running / Error | Checking the script before it runs |
| Running | The experiment is running |
| Script Error | The script stopped with an error |
| Execution Error | The run failed before or around the script |
| Registration In Progress / Error | Saving the RAI |
| Registered | Finished; the RAI PID is available |
If everything succeeds, the experiment is registered and gets a RAI. That RAI points to the datasets and script that were used. A SHA-256 hash of the results is stored with the RAI. The experiment details page shows the PID and the results.
The experiment owner can allow other registered users to download the results.
If a dataset requires result approval, the dataset owner must approve the results before others can access them. See Experiment result approvals.
