In the automated_client.py script, we search for the minimum of the Rosenbrock function, which occurs at (1,1). We are pretending that this is the result of running some computational simulation. We start with 10 data points, which form a Latin Hypercube over the bounded space. We then call the service, which calculates the point with the maximial EI (Expected Improvement). We then evaluate the rosenbrock at this point, giving us our 11th data point.

This process repeats until we sample 15 more points (representing a limited budget of computational time/runs).

Overall, this represents an automated workflow: The experiment/simulation/whatever that produces results is connected to INTERSECT. Points can be sampled automatically without human intervention.

The automated_client written here optimizes the 2D Rosenbrock function by default. To optimize functions of higher dimensions, the following changes must be made manually:

1. Update the rosenbrock function to accept additional variable(s). For example, the rosenbrock function in 3D is 100*(x1-x0**2)**2 + (1-x0)**2 + 100*(x2-x1**2)**2 + (1-x1)**2 with an optimum at (1,1,1).
2. Append a list of form [var_lower_bound, var_upper_bound] to self.bounds corresponding to each additional variable.
3. Uncomment and use the LHS code in the class initialization (instead of the explicitly written self.dataset_x) to generate initial samples in the desired dimensions that are scaled according to corresponding variable bounds.
4. Adjust surrogate meshsize (self.meshgridsize) for computational performance. A 100x100 meshgrid in 2D would become a less tractable 100x100x100 meshgrid in 3D. A mesh with a total of ~1e4 points (i.e., 100x100 or 20x20x20) will perform rapidly.
5. (Optional) Update the graph(self) function to represent the surrogate and tested points in more than 2 dimensions. If the dimension is not equal to 2, the graph will save an image with the message "Number of dimensions is not equal to two - Bayesian Optimization plot is not available. Add plotting to the graph(self) function in automated_client.py to generate a custom plot."

First, the service must be started first in a dedicated terminal by running the launch_service.py script. This should print the following: INFO:intersect-sdk:Service is starting up INFO:intersect-sdk:Service startup complete.

Second, you must also start up the mock_rosenbrock_service.py script.

Then, the client can be run in a separate terminal by running the automated_client.py script.