IRBS

This file contains:

• an overview of the code,
• a set-up guide to install it,
• a guide to run the complete project.

Overview

This code is developed in order to do a Bayesian Optimization to search for the optimal points to discover the physical specifications of a system that will lead to the desired results. From now on, the physical specifications of a system that are controlled by the user are called features, while the variables that are the outputs from the experiment are called targets.

Since the values from the targets cannot be known from the features before conducting an expensive experiment, the goal of this code is to construct a surrogate model that will aim to predict the value of the targets without running the experiment. This estimated value will help to guide the search in order to find interest targets points without running the full experiment. However, since we are interested in the real value of the targets (referred often as ground truth), after selecting some points, we will conduct the experiments and improve our surrogate model for the next cycle. This experiments will be conducted through the simulator OD, already included on this package. If in need to add another simulator, please, use the same logic existing inside models:

• a wrapper for running the simulator inside one file,
• all the auxiliary functions necessary to run the simulator on one file.

At the end of this experiment, the user should obtain more point of interest then if only running a Lattice Hypercube Sampling (LHS).

Set up

Rules and guidelines

In order to get the best out of the template:

• Don't remove any lines from the .gitignore file we provide
• Don't commit data to your repository
• Don't commit any credentials or your local configuration to your repository. Keep all your credentials and local configuration in conf/local/

Install dependencies

• When using pip (recommended) The dependencies are declared in src/requirements.txt.
  To install them, we suggest to create a virtual environment and then install them. When running with Python version than 3.6 run (from this folder):

  python -m venv venv
  . venv/bin/activate
  pip install -r src/requirements.txt

• When using pip (recommended only if your OS is Windows)

  conda create -n venv pip
  conda actiate venv
  pip install -r src/requirements.txt

Run project

In order to run this project is necessary that you have an initial file, stored in the data folder inside this project, and to add it to the file in conf/base/catalog.yml under the name initial_data, which is the first entry on the file. You can change the location of all files, but make sure that those are inside data. In order to give to the user an idea of how this file should look like, there is an initial file included in this repository, located on the direction written on the catalog.

The code is divided into 2 steps:

2. Run IRBS (Interest Region Bayesian Sampling) experiment : This pipeline will conduct the Bayesian Optimization in order to increase the initial dataset with interest points, while improving the surrogate model. It will output a DB with several samples.
3. Perform analysis on different experiments : This pipeline contains plots to analyze results, it's created to compare several experiments at the same time (described on conf/base/parameters/metrics.json).

Common parameters

There are some parameters shared between all pipelines, these are stored in conf/base/parameters.json, and they are:

• paths: Some programs need paths and store files bypassing the kedro structure,
• features: Physical variables known, that limit the design space,
• targets: Physical variables of interest, which cannot be fixed but are to be known after simulating,
• additional_values: Other values of interest that the user wishes to keep track (but not use).
• interest_region: Boundaries of the targets that will define the region of interest,

The parameters that only belong to one pipeline are located in conf/base/parameters/ and are named after each pipeline. The description of each parameter is contained in the README.md file located inside each pipeline in src/sampler/pipelines/.

Environments

In case some parameters need to be modified, they could be stored in a new folder on conf. This new folder should copy the structure on base, and should only contain the parameters or catalog entries that are modified (since all other values will be inherited from base). This new configuration can be run by adding --env new_env_folder_name to the command to run the pipeline.

2. Run IRBS (Interest Region Bayesian Sampling) experiment

kedro run --pipeline irbs --env base

3. Perform metrics

kedro run --pipeline metrics --env base