G-PlanET
RoboSense-methods
Installation
conda create -n robosense python=3.8
conda activate robosense
pip install -r requirements.txt
We should install transformers, torch-scatter, nltk, sacrebleu for Tapex model.
Noted that you should use the command pip install datasets "dill<0.3.5" to install datasets and dill issue link
File Structure
The data folder:
Included is the dataset we provide for G-PlanET, and the code for convert them to the format in iterative training and tapex training.
The main folder:
Includes all the code used in the main experiment
The script folder:
Includes all the script used in the main experiment
Use
bash script/data_init.sh
to generate the data for iterative and tapex
when using the scripts to train or test, fill in your data path and model path in the script
In our experiments, we use facebook/bart-base and facebook/bart-large for initization
The evaluation folder:
The process needed to run our test framework and its code
For more details see the readme in the folders