efficient Join Order selection learninG with Graph-basEd Representation (JOGGER) is an efficient optimizer for solving the Join order Selection(JOS) problem. It utilizes the curriculum learning, reinforcement learning and a tailored-tree-based attention module to generate query plan.
Here we have listed the most important parameters.
-ed
tree embedding dimension-gd
graph embedding dimension-b
batch size-e
episode--model
model : DQ,RTOS,JOGGER(proposed method),JOGGER w/o CLO,JOGGER w/o CTRL&CLO-f
dataset-lr
learning rate-ga
gamma-wd
weight decay
- Python 3.7
- Pytorch 1.7
- psqlparse
- deepwalk 1.0.3
- Download JOB dataset from https://github.com/gregrahn/join-order-benchmark
- Add JOB queries in the Directory: JOGGER/code/agents/queries/crossval_sens/IMDB_data.txt
- Run
encode_table.py
to build the adjacent matrix to reflect the primary-foreign key relationships - Generate the table embedding matrix according to the adjacent matrix by the deepwalk package of Python
- Run
train_JOGGER_main.py
to optimize the model
- Download TPC-H from http://www.tpc.org/tpc_documents_current_versions/current_specifications.asp
- Generate TPC-H queries from 22 templates
- Add TPC-H queries in the Directory: JOGGER/code/agents/queries/crossval_sens/TPCH_data.txt
- Run
encode_table.py
to build the adjacent matrix to reflect the primary-foreign key relationships - Generate the table embedding matrix according to the adjacent matrix by the deepwalk package of Python
- Run
train_JOGGER_main.py
to optimize the model