An implementation of ReJOIN: a learned join ordering optimizer, as described in the following papers:
- Rejoin: Hands-Free Query Optimizer through Deep Learning - Ryan Marcus & Olga Papaemmanouil
- Deep Reinforcement Learning for Joining Enumaration - Ryan Marcus & Olga Papaemmanouil
- Train target group 4 for 200 episodes
sudo python3 main.py -e 200 -g 1 -tg 4 -se 100 -s ./saved_model/group4-200/
Now the plots are in ./outputs folder (default) and the model in ./saved_model/
-
Restore saved model and test group 4
sudo python3 main.py -e 3 -g 1 -tg 4 -r ./saved_model/group4-200/ --testing -o ./outputs/testing/ -
Restore saved model and keep training on group 5 for 500 episodes
sudo python3 main.py -e 200 -g 1 -tg 5 -se 100 -r ./saved_model/group4-200/ -s ./saved_model/group5-500/ -
Execute a single query
python main.py --query 3a --episodes 150
-
Agent configuration file
"-a", "--agent-config"
default="config/ppo.json" -
Network specification file
"-n", "--network-spec"
default="config/complex-network.json" -
Number of episodes
"-e", "--episodes"
default=800 -
Total groups of different number of relations
"-g", "--groups"
default=1 -
Run specific group
"-tg", "--target_group"
default=5 -
Incremental Mode
"-m", "--mode"
default="round" -
Maximum number of timesteps per episode
"-ti", "--max-timesteps"
default=20 -
Run specific query
"-q", "--query"
default="" -
Save agent to this dir
"-s", "--save_agent" -
Restore Agent from this dir
"-r", "--restore_agent" -
Test agent without learning (use deterministic) "-t", "--testing"
action="store_true"
default=False -
Order queries by relations_num
"-all", "--run_all"
default=False -
Save agent every x episodes
"-se", "--save-episodes"
default=100 -
Select phase (1 or 2)
"-p", "--phase"
default=1