• Data Structure
• Data Process
  • Install
  • Init Client Profile
• Quickly Start Chatting with Client Agent

ECAS-DATASET
├─client                                # All generated data of Client Agents
│  ├─embodied_memory                    # Generated embodied memory of 100 Client Agnets
│  │  ├─ori_data                        # All the raw data that reflects the memory generation process
│  │  └─database.db                     # Memory stored in database
│  └─profiles                           # Generated profiles of 451 Client Agnets

1. Starting a virtual environment with anaconda:

conda create -n ecas python=3.11.0

2. Then install requirements for initing client and chatting with client agent.

pip install -r requirements.txt

To init clients, remove dulplicates and assign names to candidates, the related three Python scripts: client_init.py, init_method.py and names.py are used for processing candidate data from D4 dataset.

1. Due to the privacy of real patient data, you need get REAL data from D4 dataset website, and and put them in the ./utils/D4 folder.

2. Client initialization and duplicate removal.

   • Initializes candidate data by generating folders and saving dialogue and profile information from D4 dataset.
   • Removes duplicate candidate entries and moves non-duplicates to a new folder.
   • Assign Chinese and English names to the non duplicate candidates.
   • Copy the D4 raw data to full_portrait.json in each folder in ./client/proofiles,
   • To get the processed data, You just need run:

   python client_init.py
   

By using our generated embodied memory, you can select one of 100 clients's profile and memory to chat.

1. You need set the api_key in ./utils/utils.py (we use the OpenAI models).

2. You can use the --cn setting to select a client agent in the /client/embodied_memory/ori_data path for answering, for example (--cn parameter defaults to test0)

# use default
python question_answer.py

# use parameter
python question_answer.py --cn test0