# View help
python main.py --help

# Run dataset with custom query type
python main.py --dataset lcr --query-type SFW

Data should be saved in JSON format. Text is directly written to JSON files, while other multimodal data saves paths. Column names can be modified as needed.

{
    "id": "1",
    "description": "your-text",
    "image": "your-path-to-image"
}

Queries should be saved as query/{dataset}/{query-type}/x.sql

• --dataset, -d: Specify the dataset to run
• --all, -a: Run all datasets
• --query-type, -q: Override query type

Main UQE configuration parameters:

• USE_BART: Whether to use BART model
• BATCH_SIZE: Batch size
• BUDGET: Budget limit
• AGGR_STRATEGY: Aggregation strategy
• N_CENTROIDS: Number of clustering centroids
• N_ITER: Number of iterations
• GROUP_EXTRACT_SAMPLE_RATIO: Group extraction sampling ratio
• AGGR_CLUSTER_SAMPLE_RATIO: Aggregation cluster sampling ratio

OpenAI connection configuration:

• MODEL: Selected OpenAI model
• OPENAI_KEY: OpenAI API key
• BASE_URL: OpenAI base URL

• Results are saved in result/{dataset_name}/{query_type}/{timestamp}/ directory
• Each query result is saved as a CSV file