Using the Jupyter notebook

  1. On the Overview page, click Create new.
  2. Select Open Notebook.
  3. Specify a name for the custom algorithm.
  4. Optionally, specify a description.
  5. Save the custom algorithm to launch the Jupyter notebook.
  6. Select the created notebook on the Notebook Dashboard to open the notebook editor.
  7. To define your dataset, open the Datasets side tray to load samples of the datasets and make them available in the notebook.
    The Datasets tray serves as a dataset samples manager, where you can see a list of all available datasets. You can load and delete the samples for the notebook which you are currently working.

    When at least one dataset sample is loaded, a new folder called Datasets is created on the Notebook Dashboard page. Inside this folder, each loaded sample has a sub-folder named after the dataset UUID. Open the sub-folder to see the sample file name with a .csv extension.

    To access the dataset sample within the notebook, run a python code that references to the .csv file from the above-mentioned Dataset folder.

  8. To see a full list of available libraries, run a code line in the notebook. For example, !pip freeze or !pip list. A number of library packages are pre-uploaded and available for usage.
  9. Use the Playbook tab to prepare the custom code in a Jupyter notebook.
  10. Paste your train and predictor code in the designated tabs.
    When the train and predictor code is ready, you can switch to the designated tabs to paste the code in the code editors.
    Note: The editors are not interactive and the code is not validated. Ensure that you have the correct code posted in the train and predictor tabs.
  11. Browse to the Hyperparameters tab and select the hyperparameter definition file. The file must be in .csv format. Optionally, you can add or modify the hyperparameters directly in the grid.
    The hyperparameters can be passed in a hyperparameters.json file, as a JSON-formatted dictionary of hyperparameter names to values, or by manual input.
  12. Click Save.
  13. Click Package Files to package your custom algorithm files. A green check mark is displayed when the files are successfully packaged..
    The Package Files button is active after you provide content in the Train and Predictor tabs. Hyperparameters are optional.
    Note: The content from the Playbook tab is not part of the packaging.
  14. Click Deploy to deploy the custom algorithm.
    The status is displayed as Ready when the custom algorithm is available in the quest for training of the model.