Decision Optimization Model Builder now in Open Beta on Watson Studio Cloud
Building Decision Optimization models may require some functionality which is out of the scope of Jupyter notebooks, such as the ability to work with multiple formulations and/or data scenarios, to work with multiple files models, or to formulate models in a dedicated language like OPL. For that, a dedicated Decision Optimization Model Builder is now available in Open Beta in Watson Studio.
Why a Decision Optimization model builder?
Earlier this year, we announced the general availability of Decision Optimization support in Jupyter notebooks in Watson Studio. Notebooks are a practical way to create and combine Machine Learning and Decision Optimization models in a common environment.
However, sometimes, some optimization users may like to have some specific features which are helpful for Decision Optimization model development.
The Model Builder allows to create models in python but also in OPL or using a Modeling Assistant. You can create different scenarios, each one with different model formulation or data so that you can easily compare how they perform. A Visualization section allows creating widgets and layouts to easily share the outcome of the model to stakeholders and validate the models. Models can still be deployed to WML exactly as previously shown here.
How to start?
This functionality has been added to Watson Studio on the public cloud. If not done yet, you will need to create an IBM id and start a trial Watson Studio instance.
The entry point is to “Add a Decision Optimization model” in Watson Studio from the project view as follow:
In the next screen, you will choose a name for your model and will choose which WML instance will be used to run the models during model development. If you don’t have an instance, then create one in IBM cloud. Both services offer trials including free monthly hours so that you can easily try them.
The created Decision Optimization models are later listed in projects as any other assets:
How to use the model builder?
The model builder eases the development, debugging, tuning and validation of optimization models using 3 simple steps:
1- Import and prepare data:
In this step, you can import data from your project into any of your scenarios. You can preview your data, and modify it.
2- Formulate and run the model:
The model builder offers different ways to formulate your model, with Python, OPL or using a Modeling Assistant.
OPL and Python models can be imported from local files and Python models can even be imported from notebooks.
OPL and Python models can be using multiple files.
The modeling assistant helps you formulate a decision optimization model using natural language based on the structure of your data and some selected application domain (e.g. resource assignment, scheduling, etc).
You can see some example of the use of the Modeling Assistant in the following story about explanations.
From the ‘run model’ step, you can try your model running it. The execution will occur in the WML instance that you have selected. Some feedback is provided during the execution so that you can stop the execution at any time, for example after an acceptable feasible solution has been found.
3. Explore solution:
The last step allows previewing solution, KPIs, objective, conflicts and/or relaxations (as available).
What are the benefits of scenarios and visualization?
You also benefit from visualization (with multiple types of table and chart widgets), and the ability to work with multiple scenarios.
With Decision Optimization, (and unlike with Machine Learning) model validation is more like a manual process where you will create different scenarios, with different formulations of your model and apply them to different versions of your data. Then a subject matter expert will validate, looking at graphical representations of the solution that models are valid. Otherwise, some missing or incorrect constraints or objectives will be identified that need to be added into the model formulation. The incremental process ends when the model is complete and can be deployed to production.
Where to get more information?
You might first want to look at online documentation covering the model builder.
In particular, one way to quick start with this new functionality is to use the samples that can be created using the ‘from file’ tab of the Decision Optimization model builder creation screen. Refer to this documentation on how to start with these examples.
Don’t hesitate to ask if you have any question or problem using this beta.
LinkedIn: https://www.linkedin.com/in/alain-chabrier-5430656/
Twitter : https://twitter.com/AlainChabrier
