The UTexas system for the DARPA AIDA project (TA3)

Data Preparation

Prepare an input directory <input_dir> with TA2 KBs, each in a sub-directory named by the TA2 run name. For example:

<input_dir>
├── GAIA_TA1.Colorado_TA2
│   └── NIST
│       └── GAIA_TA1.Colorado_TA2.ttl
└── OPERA_TA1.OPERA_TA2
    └── NIST
        └── OPERA_TA1.OPERA_TA2.ttl

Prepare a SIN directory <sin_dir> with Statement of Information Need (SIN) xml files. For example:

<sin_dir>
├── T201.xml
├── T202.xml
└── T203.xml

Local Experiments

Dependencies

(Recommended) Create a virtual environment with either virtualenv or Conda.
Install PyTorch: https://pytorch.org/get-started/locally/.
Install other python dependencies by pip install -r requirements.txt.

Download the Latest Model Checkpoints

./scripts/download_models.sh --force

Run Pipeline

INPUT=<input_dir> SIN=<sin_dir> OUTPUT=<output_dir> \
    ./scripts/run_simple.sh <TA2_run_name> <TA3_run_name> [optional_args]

<input_dir> and <sin_dir> are as described in the Date Preparation section.
<output_dir> is an output directory to write intermediate results and final hypotheses.
<TA2_run_name> is the full TA2 run name, for example, GAIA_TA1.Colorado_TA2.
<TA3_run_name> is the TA3 run name to be appended, for example, UTexas_1.
[optional_args] include:
- --num_hyps <NUM_HYPS>: number of hypotheses to produce for each SIN, default = 50
- --max_num_hops <MAX_NUM_HOPS>: maximum number of hops to expand from a cluster seed to extract a subgraph, default = 5
- --min_num_eres <MIN_NUM_ERES>: minimum number of EREs to stop subgraph expansion, default = 100
- --min_num_stmts <MIN_NUM_STMTS>: minimum number of statements to stop subgraph expansion, default = 200
- --coref_compress: when specified, first compress ERE coreference on the input TA2 KB
- --plaus_rerank: when specified, use a plausibility classifier to rerank hypothesis seeds
- --device: which CUDA device to use for the neural module, default = -1 (CPU)
- --sin_id_prefix <SIN_ID_PREFIX>: the prefix of SIN IDs to use in naming the final hypotheses, default = AIDA_M36_TA3
- --force: if specified, overwrite existing output files without warning

After execution, the final hypotheses would be written to <output_dir>/<TA2_run_name>.<TA3_run_name>/, for example, <output_dir>/GAIA_TA1.Colorado_TA2.UTexas_1/. The intermediate files would be written to <output_dir>/WORKING/<TA2_run_name>.<TA3_run_name>/.

Docker

Alternatively, you can run the pipeline with our docker image without configuring dependencies.
The latest version of docker image (v0.4.1) can be found at: https://hub.docker.com/repository/docker/pxch/aida-utexas.
To run the docker image:

docker run \
    -v <input_dir>:/input:ro \
    -v <sin_dir>:/sin:ro \
    -v <output_dir>:/output \
    -e INPUT=/input \
    -e SIN=/sin \
    -e OUTPUT=/output \
    pxch/aida-utexas:0.4.1 \
    <TA2_run_name> <TA3_run_name> [optional_args]

See the Run Pipeline section for definitions of <input_dir>, <sin_dir>, <output_dir>, <TA2_run_name>, <TA3_run_name>, and [optional_args].