1. Clone the repository
2. Create a conda environment.
3. Activate the conda environment and then install the required packages using pip install -r requirements.txt
4. Make sure to activate the conda environment before running any scripts/code.

The best model is saved in the src/.models directory. To run the model on your custom equations, run the following command:

python src/symrl/run_gui.py

This will open a GUI (as shown below) where you can input your custom equations and get the results.

Warning: The GUI is only tested on Windows. It may not work on other platforms.

One can also run the best trained model on the test data by running the following command:

python src/symrl/run_inference_on_test_data.py

Lookup the folder src/.logs/_test__term_var_const_gtr_cnt_lin__td__gr for the results of the test run. The results will be dumped in a folder with timestamp as the folder name. Lookup for the file replay.log it will detailed contain the results of the test run for each of the test equations. policy.log will contain a human readable summary of the test run. summary.log will contain the summary of the test run with some aggregated statistics.

To train the model with the best configuration, run the following command:

python src/symrl/run_train_with_best_config.py

Lookup the folder src/.logs/term_var_const_gtr_cnt_train__simpl_term_var_const_count_lin__td__eps for the results of the test run. The model along with the logs will be saved in the folder with timestamp as the folder name. Individual checkpoints will be saved in the src/.logs/term_var_const_gtr_cnt_train__simpl_term_var_const_count_lin__td__eps/<timestamp>/model/ckpt folder. The training logs will be dumped in the src/.logs/term_var_const_gtr_cnt_train__simpl_term_var_const_count_lin__td__eps/<timestamp>/ folder with the name policy.log, replay.log and summary.log.

To run the exhaustive training and ablation study, run the following command:

./src/symrl/experiments.sh

Warning: This can take 1-2 days to complete depending on the hardware configuration.

To generate the report (i.e. plots and table summary) after running the exhaustive training and ablation study, run the following command:

python src/symrl/analysis.py --analysis_json "<path-to-analysis-json>" --plot_folder "<path-to-output-dir>" --plot_name "<name-of-the-plot>"

You will have to create the analysis json file which might look like:

[
    {
        "name": "TVCC w Comparison",
        "folder": "src/.logs/100_25_test__simpl_term_var_const_count_lin__td__gr_eval/20240418_064630"
    },
    {
        "name": "TVCC",
        "folder": "src/.logs/100_25_test__term_var_const_count_lin__td__gr_eval/20240418_061200"
    },
    {
        "name": "TVC w Comparison",
        "folder": "src/.logs/100_25_test__op_var_count_lin__td__gr_eval/20240418_061200"
    },
    {
        "name": "TC",
        "folder": "src/.logs/100_25_test__op_count_lin__td__gr_eval/20240418_061200"
    },
    {
        "name": "Hand-Crafted Strategy",
        "folder": "src/.logs/human_test/20240421_153646"
    },
    {
        "name": "Random Strategy",
        "folder": "src/.logs/random_test/20240422_001813"
    }
]

Note: Do remember to replace the paths with the actual paths of the folders (usually you just have to replace the timestamp in the path).