An unlimited amount of training data for machine learning can be created and stored using Stateshaper. The ability for the engine to derive synthetic data by tokenizing its numeric output allows for a wide range of training data values to be used. Each test can be stored and re-created at any time from the small seed formats seen to the right of the screen.
If the data variations created aren't good enough, output can be adjusted in the plugin file by using the current token as a base to derive test values from. This particular example shows how Stateshaper can be used to run road simulations that help train the AI in self-driving cars. Theoretically, all possible test scenarios can be created based on how output of the program is structured. If needed, the parameter values for the main class and corresponding plugin file can be modified for a particular use.
Once run, these tests can be stored using almost no space. Any simulation can be revisited at any time. Consider that one prototype for a self-driving car may have several versions. Each version created can have millions of possible AI training sessions conducted before the car begins testing on the road. The data needed for this can take up many terabytes of space.
Storing this data can be important for many reasons such as further research, version comparison, and regulatory reasons to name a few. Using Stateshaper in this case can reduce database related costs in this instance by over 99%. This includes storage, bandwidth and electricity consumption. This logic in this demo can also be used in many other applications that require ML Training.
Stateshaper is currently available as a Python package and Github repository.
ML Training is only one of the many uses for this program. There are other demos listed in the project's documentation. Other uses can include, but are not limited to, smart home scheduling, gaming NPC behavior, content generation, graphic assets and store inventories.