Creating AutoML Experiment
A Data Scientist can create various Experiments based on specified algorithms.
Last updated
A Data Scientist can create various Experiments based on specified algorithms.
Last updated
There can be different types of Experiments based on the algorithm type specified. In the DS Lab module, we currently support Classification, Regression, and Forecasting.
A Classification experiment can be created for discrete data when the user wants to predict one of the several categories.
A Regression experiment can be created for continuous numeric values.
A Forecasting experiment can be created to predict future values based on historical data.
Please Note:
AutoML experiments are running as Jobs and a new Job will be allocated for each experiment created in the AutoML tab.
Jobs will spin up once the Experiment is created and after models are trained and ready, it will get killed automatically.
Creating an Experiment is a two-step process that involves configuration and selection of the algorithm type as steps.
A user can create a supervised learning (data science) experiment by choosing the Create Experiment option.
Please Note: The Create Experiment icon is provided on the Dataset List page under the Dataset tab of a Repo Sync Data Science Project.
Navigate to the Data List page.
Select a Dataset from the list.
Click the Create Experiment icon.
The Configure tab opens (by default) while selecting the Create Experiment option.
Provide the following information:
Provide a name for the experiment.
Provide Description (optional).
Select a Target Column.
Select a Data Preparation from the drop-down menu.
Use the checkbox to select a Data Preparation from the displayed drop-down.
Select columns that need to be excluded from the experiment.
Use the checkbox to select a field to be excluded from the experiment.
Please Note: The selected fields will not be considered while training the Auto ML model experiment.
Click the Next option.
The user gets redirected to the Select Experiment Type tab.
Select a prediction model using the checkbox.
Based on the selected experiment type a validation notification message appears.
Click the Done option.
A notification message appears.
The user is redirected to the AutoML list page.
The newly created experiment gets added to the list with Status mentioned as Started.
The Status tab indicates various phases of the experiments/model training. The different phases for an experiment are as given below:
The newly created experiment gets Started status. It is the first status when a new experiment is created.
Another notification message appears to inform the user that the model training has started. The same is indicated through the Status column of the model. The Status for such models will be Running.
After the experiment is completed, a notification message appears stating that the model trained. The Status for a trained model will be indicated as Completed.
Please Note: The unsuccessful experiments are indicated as Failed under the status. The View Report is mentioned in red color for the Failed experiments.
