Merge branch 'master' of...

#!/usr/bin/env bash

#SBATCH --job-name=embedding_types

#SBATCH --partition=gpu_4

#SBATCH --mem=16G

#SBATCH --gres=gpu:1

#SBATCH --cpus-per-task=1

#SBATCH --time=3:00:00

#SBATCH --output=embedding_type_study.out

#SBATCH --mail-user=hoepfl@cl.uni-heidelberg.de

#SBATCH --mail-type=ALL

#SBATCH --ntasks=1

source ~/.bashrc

conda activate diac_lm

SEED_DICT=~/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/clean_dict_sme_nob/clean_single_seed_dict.txt

TEST_DICT_SINGLE=~/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/clean_dict_sme_nob/clean_single_dev_dict.txt

TEST_DICT_MULTI=~/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/clean_dict_sme_nob/clean_multi_dev_dict.txt

# Clean NOB + passende SME-Embeddings:

PAIRS=(

  "combined_all_fasttext_300_cbow.vec|clean_corpus_nob_fasttext_300_cbow.vec"

  "corpus_sme_fasttext_300_skip_gram.vec|clean_corpus_nob_fasttext_300_skip_gram.vec"

  "corpus_sme_glove_300_vectors.vec|clean_corpus_nob_glove_300_vocab.txt"

  "corpus_sme_word2vec_300.vec|clean_corpus_nob_word2vec_300.vec"

)

for pair in "${PAIRS[@]}"; do

    SME_FILE=$(echo "$pair" | cut -d'|' -f1)

    NOB_FILE=$(echo "$pair" | cut -d'|' -f2)

    echo "-----------------------------------------"

    echo "Mapping: $SME_FILE -> $NOB_FILE"

    echo "-----------------------------------------"

    SME_PATH="$HOME/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/embeddings/${SME_FILE}"

    NOB_PATH="$HOME/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/embeddings/${NOB_FILE}"

    # 1) Map Embeddings (semi-supervised)

    srun python ~/diac_lm_n/diachronic-llms-automatic-dictionary-induction/vecmap/map_embeddings.py \

         --semi_supervised \

         "$SEED_DICT" \

         "$SME_PATH" \

         "$NOB_PATH" \

         ~/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/embeddings/sme_mapped_temp \

         ~/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/embeddings/nob_mapped_temp \

         --cuda

    # 2) Evaluate Single

    srun python ~/diac_lm_n/diachronic-llms-automatic-dictionary-induction/vecmap/eval_translation.py \

         ~/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/embeddings/sme_mapped_temp \

         ~/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/embeddings/nob_mapped_temp \

         -d "$TEST_DICT_SINGLE" \

         --cuda --retrieval csls \

         > "embedding_type_study/record_file_single_$(basename "${SME_FILE}" .vec).txt"

    # 3) Evaluate Multi

    srun python ~/diac_lm_n/diachronic-llms-automatic-dictionary-induction/vecmap/eval_translation.py \

         ~/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/embeddings/sme_mapped_temp \

         ~/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/embeddings/nob_mapped_temp \

         -d "$TEST_DICT_MULTI" \

         --cuda --retrieval csls \

         > "embedding_type_study/record_file_multi_$(basename "${SME_FILE}" .vec).txt"

    echo "Fertig mit $SME_FILE -> $NOB_FILE"

    echo

done

# Konfuzius sagt:

# "Wer die besten Embeddings sucht, muss erst die schlechtesten vergessen."