Releases: osoblanco/Query-Answering-In-Knowledge-Space
Releases · osoblanco/Query-Answering-In-Knowledge-Space
Valid and test queries with 1p queries for Q2B evaluation
To test a model on 1p queries, specify 1_1 as the query type. For example:
python continuous.py --model_path models/FB15k-237-model-rank-500-epoch-100-1602506111.pt --dataset FB15k-237 --dataset_mode test --chain_type 1_1
valid+test FB15k, FB15k-237, NELL triples from Q2B
These triples were obtained from the Q2B datasets and can be used to train and evaluate for the case of 1p queries.
NELL triples for KBC, and queries
Download NELL.zip to kbc/data and decompress.
With the root of the repository as the current directory, train ComplEx with
python -m kbc.learn kbc/data/NELL --model ComplEx --max_epochs 50 --model_save_schedule 10 --valid 1 --reg 1e-2 --batch_size 1024 --rank 500
0.6: Merge pull request #1 from dfdazac/master
Sample queries that rely on at least 1 test triple
Converting Q2B data to RCQA format
This release includes the notebook for the conversion of the data from the Q2B (https://github.com/hyren/query2box) format tot he one used in our implementation, along with the converted files.
In order to use the notebook either put it in the /data of the query2box repo or kindly change the paths designated in the code.