If you also use XLING data and save the dataset in the same directories as we did, you could directly run the code without modifying anything; you might only need to pay a little attention to (c) and (d) below. When using customised data, you would also need to modify the input/output directories specified in (a) and (b) below.
(a) In ./C1/run_all.py, set the directories for (1) pretrained static word embeddings for source and target languages respectively; (2) test dictionary; (3) training dictionary; (4) saving C1-aligned CLWEs.
(b) In ./C2/run_all.py, set the directories for (1) saving hard negative pairs extracted from C1; (2) saving fine-tuned multilingual LMs; (3) pretrained multilingual LMs; (3) some hyper parameter values for C2; (4) pretrained static word embeddings for source and target languages; (5) test dictionary; and (6) training dictionary.
(c) To run supervised (5k seed pairs) and semi-supervised (1k seed pairs) BLI tasks, please set the size of the training dictionary to '5k' or '1k' respectively in both ./C1/run_all.py and ./C2/run_all.py.
(d) To save C1-aligned embeddings for C2's use, please set args.save_aligned_we = True in ./C1/src/main.py. If you run the code for the first time, you might also need to mkdir directories for saving C1-aligned CLWEs, hard negative pairs extracted for C2, and C2 model checkpints, respectively.