3x3_metasurface

This is the codebase for building the custom dataset used for the study presented in the publication, Time-series neural networks to predict electromagnetic wave propagation.

---

Running from a local Docker container

A library of 3x3 pillar radii between 75 nm and 250 nm is generated using generate_radii.py. This generates a file called neighbors_library_allrandom.pkl. This file stores a numpy array with len(5000). Each index contains a list of 9 radii.

Running main.py allows the user to either generate data or reduce the data into volumes, as presented in the paper linked above.

1. Generate data. To run a single simulation:

   • Set the config file parameter, task to 0.
   • Create an instance of the Docker image, v3_lightning.
   • From /develop/code/3x3_metasurface, run:

     mpirun --allow-run-as-root -np {num_cpu_cores} python3 main.py -config configs/config.yaml -idx {neighbors library index}
     

2. Reduce data.

   • Set the config file parameter, task to 0.
   • Running the same Docker container in (1), from /develop/code/3x3_metasurface, run:

     python3 reduce_data.py
     

---

Using Kubernetes

1. Generate data. To launch many simulations:

• Set the config file parameters.
  • task : 0
  • deployment_mode : 1
  • kube.datagen_job.num_sims : 1500
  • kube.datagen_job.start_group_id : 0
  • kube.datagen_job.num_parallel_ops : {your preference}
• Running an instance of the Docker image, kubernetes launcher, navigate to kubernetes/gen_data/ and run

  python3 launch_jobs.py
  

2. Reduce data.

• Set the config file parameters.
  • task : 1
  • deployment_mode : 1
• From kubernetes/reduce_data, run:

  python3 launch_jobs.py