Training Playground is a training environment which is present in each Experiment of any AI Task. It is the place where you bring your training data and build the pipeline in order to create a trained AI model. The AI model learns from historical information and then predicts answers to your question. Each run of this environment will create a new AI model. You can run the training environment multiple times and create multiple models (e.g. by bringing more data or changing the training time and running the environment again).
After configuring the Pipeline, you can train an AI model. Please check that there are no disconnected items and unconfigured blocks (with an orange warning icon).
If the Wand Platform detects any problem with creating the trained AI model, you will receive a warning in the run log. Like here – the ML task was not configured correctly:
The training might take a long time – however, you can pre-determine the training time. You can close the running log and do something else. The Wand Platform will send you an email after a successful or unsuccessful run. If the run was successful, an AI model will be created.
A successful run creates an AI model and also generates predictions, explainability, and metrics on the automatically chosen test set (a random holdout subset of the whole original dataset). A trained AI model can be deployed to Staging or Production and generate predictions and explainability for new datasets.
Retraining the AI model
You can always retrain an existing AI model in the Training Playground, e.g. adding more labeled data, changing data transformations, or using other settings of the Wand ML block. The retrained model can be re-deployed to Staging or Production. Please note that if the data schema (column names, type of AI task, or prediction key) changes, then the Wand Platform will create a new Experiment (within the same AI task) for this schema.