Machine learning workflow in Granta MI

Overview of how machine learning is implemented in Granta MI, including a glossary of key terminology used in MI Machine Learning, which app to use to complete a task, and information on the Intellegens Alchemite engine and algorithm.

Tip: The machine learning workflow in Granta MI is divided between the MI Machine Learning apps:

Use the Training app to create feature matrices and train machine learning algorithms:
1. Create a model and add Granta MI attributes and records to its Data Matrix.
2. Train a model by sending it to the machine learning engine.
  - The Alchemite engine uses a neural network algorithm, and part of the training dataset is used to test the trained model before returning it.
  - The engine runs on a dedicated machine learning server, separate from the SQL Server and main application server for Granta MI.
  - When a model is sent for training, MI Machine Learning transforms the Matrix of attributes and records into a feature matrix and provides that input to the engine.
Use the Visualize app to view your trained model's outputs and correlation matrix. You can visualize models by:
- Plotting an input feature and an output feature across the entire training range of the model, specifying fixed values for all other features (Plot model features).
- Plotting selected output features on parallel axes (View model outputs).
Use the Optimize app to solve process optimization problems with a trained model.
1. Create a Criteria Set for a trained model:
  - Specify constraints to apply to input features.
  - Choose target values (goals) for output features.
2. Send the model and Criteria Set to the engine for evaluation.
3. View the results returned by the engine, including the estimated likelihood of achieving the output feature values if those input values are true.
4. Pre-populate the Visualize app Plot model features or View model outputs tabs with your results to explore the predicted solution further.