svp-sdml

• Data (Word-in-Context)
  • AM2iCo
  • MCL-WiC
  • XL-WiC
• Data (Semantic Change Detection)
  • SemEval-2020 Task 1
  • RuShiftEval

1. Setup

1.1 Load environments

use Dockerfile and

setup.sh

1.2 Preprocess WiC datasets

we provide shell script of preprocessing

preprocess_wic.sh

AM2iCo

python3 src/preprocess_wic.py \
    --am2ico_path path/to/am2ico/lang/data.tsv \
    --filename path/to/result/lang/data.tsv

MCL-WiC

python3 src/preprocess_wic.py \
    --mclwic_data path/to/mclwic/lang/data.data \
    --mclwic_gold path/to/mclwic/lang/data.gold \
    --filename path/to/result/lang/data.tsv

XL-WiC

python3 src/preprocess_wic.py \
    --xlwic_path path/to/xlwic/lang/data.txt \
    --filename path/to/result/lang/data.tsv

1.3 Calculate vectors

python3 src/prepare_vec_wic.py \
    --data_dir path/to/wic/processed/data/dir

2. Main

2.1 Load environments

use Dockerfile or

pip install -r requirements.txt

2.2 Training

python3 src/main.py \
    --train_data_pathes path/to/wic/vec/train \
    --dev_data_pathes path/to/wic/vec/dev \
    --test_data_pathes path/to/wic/vec/test \
    --search_param

2.3 Calculate vectors

SemEval

python3 src/prepare_vec_semeval.py \
    --file_path path/to/semeval/token.txt \
    --lemma_path path/to/semeval/lemma.txt \
    --target_words_list path/to/targets.txt \
    --lang [en, de, sw, la] \
    --output_name path/to/result/pkl

RuShiftEval

python3 src/prepare_vec_rushifteval.py \
    --file_path path/to/rushifteval/data.tsv \
    --target_words_list path/to/rushifteval/targets.txt \
    --output_name path/to/result/pkl

2.4 Evaluate

python3 src/evaluate.py \
    --gold_path path/to/graded.txt \
    --model_path path/to/model.pkl \
    --vec_pathes path/to/result/pkl