Working with experiments
Load historical data into an automated machine learning experiment and train a model to analyze and predict a business problem.
You can create and edit experiments in personal or shared spaces.
Workflow
Before you create an automated machine learning experiment in Qlik Cloud Analytics, you need to have a well-defined machine learning question and a suitable training dataset available in Catalog. For more information, see Defining machine learning questions and Getting your dataset ready for training.
The following steps describe an experiment workflow.
- Create your experiment
Create a new experiment in Qlik Sense. Add it to a shared space if you want to work collaboratively.
- Configure your experiment
Select a target to make predictions on and features to support the prediction.
- Start the training
Start the training of your first experiment version.
- Review recommended models
Review how the model training went and assess the models that have been recommended to you. Use the built-in tools to analyze training summaries and embedded analytics, which offer insights at the experiment, version, and model level.
If further refinement is needed, you can configure subsequent versions of the experiment. Adjust parameters such as features and algorithms and retrain new versions of the experiment until you have a good model.
- Deploy the model
When you have a good model, it’s time to deploy it and start making predictions.
Requirements and permissions
For information about user permissions requirements for working with ML deployments and predictions, see Access controls for ML experiments.
Viewing lineage and impact analysis
Using the Lineage and Impact analysis tools in Qlik Cloud, you can analyze:
-
The origins of an ML experiment and the data used to train each model.
-
How the ML experiment and its outputs are being used in downstream content across Qlik Cloud.