FARSI Simulator Documentation

1. Clone ArchGym Repository and Create Conda Environment

Do these steps from the location where you intend to clone the ArchGym repository:

git clone https://github.com/srivatsankrishnan/oss-arch-gym.git
conda env create -f oss-arch-gym/environment.yml
conda activate arch-gym

2. Install Vizier for Collection of Agents

From the repository root oss-arch-gym run:

./install_sim.sh viz

3. Installing FARSI simulator

  • The below commands are to replace the existing Project_FARSI folder with its latest version as a submodule. The shell script also updates the conda environment dependencies required for FARSI, and installs ACME framework for reinforcement learning.

(Note: the script takes a while to run):

cd oss-arch-gym/
rm -r Project_FARSI
git rm -r --cached Project_FARSI
./install_sim.sh farsi

  • Replace the content of Project_FARSI/settings/config_cacti.py file with this:

import os

# get the base path of arch-gym
base_path = os.path.abspath(os.path.join(os.path.dirname(__file__), '../..'))


cact_bin_addr = os.path.join(base_path, "Project_FARSI/cacti_for_FARSI/cacti")

print(cact_bin_addr, os.path.exists(cact_bin_addr))

cacti_param_addr = os.path.join(base_path, "Project_FARSI/cacti_for_FARSI/farsi_gen.cfg")

print(cacti_param_addr, os.path.exists(cacti_param_addr))

cacti_data_log_file = os.path.join(base_path, "Project_FARSI/cacti_for_FARSI/data_log.csv")

print(cacti_data_log_file, os.path.exists(cacti_data_log_file))

  • In Project_FARSI/settings/config.py, replace the following line (line 276):

database_data_dir = os.path.join(home_dir, "specs", "database_data")

  • with this:

database_data_dir = os.path.join(home_dir, "oss-arch-gym", "Project_FARSI", "specs", "database_data")

Running Training Scripts

Inside sims/FARSI_sim:

  • Ant Colony Optimization: python train_aco_FARSIEnv.py

  • Bayesian Optimization: python train_bo_FARSIEnv.py.py

  • Genetic Algorithm: python train_ga_FARSIEnv.py

  • Random Walker: python train_randomwalker_FARSIEnv.py

  • Reinforcement Learning: python train_single_agent.py

  • Emukit Algorithm: python train_emukit_vizier.py

  • Grid-Search Algorithm: python train_gridsearch_vizier.py

  • Quasi-Random Algorithm: python train_quasirandom_vizier.py

  • Random-Search Algorithm: python train_randomsearch_vizier.py

Updating Hyperparameters

You can update hyperparameters of the different algorithms in their respective .py files shown above.