Skip to content

Loading Experimental Data

The Experiment Data panel in ALchemist lets you load, view, and manage your experimental results. This is a key step before training surrogate models or running active learning.

Loading Data from File

  1. Click "Load Experiments":
    In the Experiment Data panel, click the Load Experiments button.
  2. Select Your File:
    Choose a .csv file containing your experimental data. The file should have columns for each variable (matching your variable space) and an Output column for the measured result. Optionally, you can include a Noise column to specify measurement uncertainty for each point.
  3. Data Appears in the Table:
    The loaded data will be displayed in the table. If a Noise column is present, it will be used for model regularization.

Tip:
If your data columns do not match the variable names or required format, you may see an error. Make sure your CSV headers match your variable space exactly.

Adding a New Experiment Point

You can add a new experiment directly from the UI:

  1. Click "Add Point":
    Opens a dialog where you can enter values for each variable, the output, and (optionally) the noise.
  2. Fill in the Fields:
    Enter values for all variables and the output. If you know the measurement uncertainty, enter it in the Noise field.
  3. Save & Close:
    Click Save & Close to add the point to your experiment table. You can also choose to save the updated data to file and retrain the model immediately by checking the corresponding boxes.

Note:
- There may be issues with type compatibility (e.g., numbers being saved as strings). If you encounter problems, check your CSV file and ensure numeric columns are formatted correctly. - Sometimes, changes made directly in the table (tksheet widget) may not update the internal experiment data until you save or reload. Use the provided dialogs for best results.

Saving Your Data

  • Click "Save Experiments":
    Saves the current experiment table to a .csv file.
  • Tip:
    Always save your data before closing the application to avoid losing changes.

Retraining the Model

  • When adding a new point, you can check Retrain model to automatically update the surrogate model with the new data.
  • If retraining does not seem to trigger, you may need to retrain manually from the model panel.

Known Issues & Tips

  • Type Compatibility:
    Data entered via the table or add-point dialog may sometimes be interpreted as strings. If you see errors or unexpected behavior, check your data types in the CSV file.
  • Table Edits:
    Editing data directly in the table does not always update the internal experiment manager. For reliable results, use the add-point dialog or reload your data after editing.
  • Noise Column:
    The noise column is optional. If present, it should be numeric. You can toggle its visibility in the Preferences menu.

For more details on managing experiments and troubleshooting, see the rest of the workflow documentation.