Module deepcomp.util.constants

"""
Utility module holding constants that are used across different places.
Such that this is the single point in the code to adjust these constants.
"""
import os
import pathlib


# logging settings
LOG_ROUND_DIGITS = 3

# use sets for O(1) include checks
CENTRAL_ALGS = {'ppo', 'random', 'fixed', 'brute-force'}
MULTI_ALGS = {'ppo', 'greedy-best', 'greedy-all', 'dynamic'}
SUPPORTED_ALGS = CENTRAL_ALGS.union(MULTI_ALGS)
SUPPORTED_AGENTS = {'single', 'central', 'multi'}
SUPPORTED_ENVS = {'small', 'medium', 'large', 'custom'}
SUPPORTED_RENDER = {'html', 'gif', 'both', None}
SUPPORTED_SHARING = {'max-cap', 'resource-fair', 'rate-fair', 'proportional-fair'}
SUPPORTED_REWARDS = {'min', 'sum'}

# small epsilon used in denominator to avoid division by zero
EPSILON = 1e-16

# constants to tune "fairness" of proportional-fair sharing
# high alpha --> closer to max cap; high beta --> closer to resource-fair; alpha = beta = 1 is used in 3G
# actually no, alpha=1=beta converges to exactly the same allocation as resource-fair for stationary users!
# https://en.wikipedia.org/wiki/Proportionally_fair#User_prioritization
FAIR_WEIGHT_ALPHA = 1
FAIR_WEIGHT_BETA = 1

# constants regarding result and trainig files
# project root (= repo root; where the readme is) for file access
_this_dir = pathlib.Path(__file__).parent.absolute()
PROJECT_ROOT = _this_dir.parent.parent.absolute()
RESULT_DIR = os.path.join(PROJECT_ROOT, 'results')
TRAIN_DIR = os.path.join(RESULT_DIR, 'PPO')
TEST_DIR = os.path.join(RESULT_DIR, 'test')
VIDEO_DIR = os.path.join(RESULT_DIR, 'videos')
PLOT_DIR = os.path.join(RESULT_DIR, 'plots')


def create_result_dirs():
    """Create directories for saving training, testing results and videos"""
    os.makedirs(RESULT_DIR, exist_ok=True)
    os.makedirs(TRAIN_DIR, exist_ok=True)
    os.makedirs(TEST_DIR, exist_ok=True)
    os.makedirs(VIDEO_DIR, exist_ok=True)
    os.makedirs(PLOT_DIR, exist_ok=True)


create_result_dirs()