How do I run the ANDH-Full task?

[Wang1665941855]

Dear Author, how do I run the ANDH-Full task? Should I uncomment the code in the env.py file and modify the train_val_on_full parameter?

[UeFan]

Hi sorry for the late reply.

Please ignore those commented code. The evaluation for ANDH-Full task should be the same as the evaluation for ANDH with just the data being different. For the full trajectory data needed in ANDH-Full task, please check out the data I just uploaded to Huggingface: AVDN Data Repo.

[tzhhhh123]

Dear Author, thank you for your guidance. I have managed to achieve approximate results by following these steps:

I converted the downloaded CSV files to JSON format.
I replaced the split names in the dataset to align with the full mode.
I enabled the train_on_val option for testing purposes.
Additionally, I modified some keys' types and names in the CSV-to-JSON conversion script to ensure consistency with the original code.
With these adjustments, the results are close, but I wanted to confirm if there are any other recommended steps or adjustments to ensure full alignment with the intended ANDH-Full task implementation.