Commit 7c0740ef authored by chrysanthopoulou's avatar chrysanthopoulou
Browse files

upload slurm scripts

parent 7874c592

slurm_scripts/embeddings_grid_search.sh

0 → 100644
+113 −0
Original line number Diff line number Diff line 
okay, Finn can you have a look whether this looks plausibel? 





#!/usr/bin/env bash

#SBATCH --job-name=grid_dims_classymap

#SBATCH --partition=dev_gpu_4_a100

#SBATCH --mem=16G

#SBATCH --cpus-per-task=32

#SBATCH --gres=gpu:1

#SBATCH --cpus-per-task=1

#SBATCH --time=30:00

#SBATCH --output=grid_dims_classymap.out

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

#SBATCH --mail-type=ALL

#SBATCH --ntasks=1



#source ~/.bashrc

#conda activate diac_lm



# Load required modules

module load devel/cuda/12.4



# Set library path

export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH



SME_CORPUS="$HOME/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/sami_corpora/final_sme_corpus.txt"

NOB_CORPUS="$HOME/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/sme_nob_corpora/clean_corpus_nob.txt"



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



EMB_DIR="$HOME/diac_lm_n/diachronic-llms-automatic-dictionary-induction/data/embeddings"

# Save to subfolder of slurm_scipts, not embeddings since those will probably be throwaway results

ALIGN_DIR="$HOME/diac_lm_n/diachronic-llms-automatic-dictionary-induction/slurm_scripts/dim_study/aligned"

MODEL_DIR="$HOME/diac_lm_n/diachronic-llms-automatic-dictionary-induction/slurm_scripts/dim_study/classifier"

RESULTS_DIR="$HOME/diac_lm_n/diachronic-llms-automatic-dictionary-induction/slurm_scripts/dim_study/translation_candidates"





DIMS=(50 100 300 500)



for dim in "${DIMS[@]}"; do

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

  echo "Generating embeddings for dimension = $dim"

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



  # Generate FastText embeddings for SME

  srun fastText/fasttext skipgram \

    -input "$SME_CORPUS" \

    -output "${EMB_DIR}/sme_dim${dim}_skip_gram" \

    -dim "$dim" \

    -thread 64



  # Generate FastText embeddings for NOB

  srun fastText/fasttext skipgram \

    -input "$NOB_CORPUS" \

    -output "${EMB_DIR}/nob_dim${dim}_skip_gram" \

    -dim "$dim" \

    -thread 64



  # Paths to .vec version of embeddings

  SME_EMB="${EMB_DIR}/sme_dim${dim}_skip_gram.vec"

  NOB_EMB="${EMB_DIR}/nob_dim${dim}_skip_gram.vec"



  echo "Mapping embeddings for dimension = $dim using ClassyMap..."



  # ClassyMap training

  srun python3 ClassyMap/src/classymap.py \

    --train_dict "$SEED_DICT" \

    --in_src  "$SME_EMB" \

    --in_tar  "$NOB_EMB" \

    --src_lid sme \

    --tar_lid nob \

    --out_src "${ALIGN_DIR}/aligned_sme_dim${dim}.vec" \

    --out_tar "${ALIGN_DIR}/aligned_nob_dim${dim}.vec" \

    --model_filename "${MODEL_DIR}/classifier_model_dim${dim}.pkl" \

    --use_mnns_pooler 1 \

    --cuda



  # ClassyMap evaluation

  srun python3 ClassyMap/src/eval.py \

    "${ALIGN_DIR}/aligned_sme_dim${dim}.vec" \

    "${ALIGN_DIR}/aligned_nob_dim${dim}.vec" \

    --dictionary "$TEST_DICT_SINGLE" \

    --output_file "${RESULTS_DIR}/single_translation_candidates_dim${dim}.txt" \

    --src_lid sme \

    --tar_lid nob \

    --idstring "EXP_dim${dim}" \

    --retrieval csls \

    --seed 42 \

    --cuda \

    --super \

    --model "${MODEL_DIR}/classifier_model_dim${dim}.pkl"



    # ClassyMap evaluation

  srun python3 ClassyMap/src/eval.py \

    "${ALIGN_DIR}/aligned_sme_dim${dim}.vec" \

    "${ALIGN_DIR}/aligned_nob_dim${dim}.vec" \

    --dictionary "$TEST_DICT_MULTI" \

    --output_file "${RESULTS_DIR}/multi_translation_candidates_dim${dim}.txt" \

    --src_lid sme \

    --tar_lid nob \

    --idstring "EXP_dim${dim}" \

    --retrieval csls \

    --seed 42 \

    --cuda \

    --super \

    --model "${MODEL_DIR}/classifier_model_dim${dim}.pkl"



done



# Konfuzius sagt:

# "Wer vier Dimensionen testet, braucht eine fünfte für die Geduld."
 
 No newline at end of file

slurm_scripts/gen_emb.sh

0 → 100644
+36 −0
Original line number Diff line number Diff line 
#!/usr/bin/env bash

#SBATCH --job-name=grid_search

#SBATCH --partition=dev_gpu_4_a100

#SBATCH --mem=50G

#SBATCH --cpus-per-task=10

#SBATCH --gres=gpu:1

#SBATCH --time=00:15:00

#SBATCH --output=gen_sme_emb.out

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

#SBATCH --mail-type=ALL

#SBATCH --ntasks=1



#source ~/.bashrc

cd $HOME/diachronic-llms-automatic-dictionary-induction



# Load required modules

module load devel/cuda/12.4



# Set library path

export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH



# Fixed dimension

DIM=300



# Corpora paths

SME_CORPUS="$HOME/diachronic-llms-automatic-dictionary-induction/data/sami_corpora/final_sme_corpus.txt"



EMB_DIR="$HOME/diachronic-llms-automatic-dictionary-induction/data/embeddings"





srun fastText/fasttext skipgram \

    -input "$SME_CORPUS" \

    -output "final_sme_corpus_fasttext_300_skipgram" \

    -dim "$DIM" \

    -thread 64

slurm_scripts/run_grid_search.sh

+6 −6
Original line number Diff line number Diff line 
#!/usr/bin/env bash

#SBATCH --job-name=grid_search

#SBATCH --partition=gpu_4

#SBATCH --partition=gpu_8

#SBATCH --mem=50G

#SBATCH --cpus-per-task=10

#SBATCH --gres=gpu:1

#SBATCH --time=15:00:00

#SBATCH --output=gridsearch_study_lower_ressources_4.out

#SBATCH --time=00:10:00

#SBATCH --output=gridsearch_study_5_2_only.out

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

#SBATCH --mail-type=ALL

#SBATCH --ntasks=1
@@ -23,9 +23,9 @@ export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH 
DIM=300



# Grid 

MIN_COUNTS=(2 5 10)

MIN_NS=(2 3)

MAX_NS=(2 5 10)

MIN_COUNTS=(2)

MIN_NS=(3)

MAX_NS=(2)



# Corpora paths

SME_CORPUS="$HOME/diachronic-llms-automatic-dictionary-induction/data/sami_corpora/final_sme_corpus.txt"

slurm_scripts/run_grid_search_man.sh

0 → 100644
+229 −0
Original line number Diff line number Diff line 
#!/usr/bin/env bash

#SBATCH --job-name=grid_search

#SBATCH --partition=gpu_8

#SBATCH --mem=80G

#SBATCH --gres=gpu:1

#SBATCH --time=00:50:00

#SBATCH --output=best_cov_%j.out

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

#SBATCH --mail-type=ALL

#SBATCH --ntasks=1



#source ~/.bashrc

cd $HOME/diachronic-llms-automatic-dictionary-induction



# Load required modules

module load devel/cuda/12.4



# Set library path

export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH



# Fixed dimension

DIM=300



# Grid 

mc=10

#(2 5 10)

mn=3

#(2 3)

mx=5

#(2 5 10)



# Grid 

mcs=2

#(2 5 10)

mns=2

#(2 3)

mxs=10

#(2 5 10)



# Corpora paths

SME_CORPUS="$HOME/diachronic-llms-automatic-dictionary-induction/data/sami_corpora/final_sme_corpus.txt"

NOB_CORPUS="$HOME/diachronic-llms-automatic-dictionary-induction/data/nob_corpora/clean_corpus_nob.txt"



# Dictionaries

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

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

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



# Output directories - not saved to data/embeddings since those will likely by throwaway embeddings

EMB_DIR="$HOME/diachronic-llms-automatic-dictionary-induction/slurm_scripts/ngram_study/embeddings"

ALIGN_DIR="$HOME/diachronic-llms-automatic-dictionary-induction/slurm_scripts/ngram_study/aligned"

MODEL_DIR="$HOME/diachronic-llms-automatic-dictionary-induction/slurm_scripts/ngram_study/classifier"

RESULTS_DIR="$HOME/diachronic-llms-automatic-dictionary-induction/slurm_scripts/ngram_study/translation_candidates"



# Standard embeddings

SME_EMB_STANDARD="$HOME/diachronic-llms-automatic-dictionary-induction/data/embeddings/combined_all_fasttext_300_skip_gram.vec"

NOB_EMB_STANDARD="$HOME/diachronic-llms-automatic-dictionary-induction/data/embeddings/clean_corpus_nob_fasttext_300_skip_gram.vec"







# echo "---------------------------------------------"

# echo "testing SME = Standard, NOB = minCount=$mc | minn=$mn | maxn=$mx"

# echo "---------------------------------------------"



# SME_EMB=$SME_EMB_STANDARD

# NOB_EMB="${EMB_DIR}/nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}.vec"



# # ClassyMap training

# srun python3 ClassyMap/src/classymap.py \

#   --train_dict "$SEED_DICT" \

#   --in_src  "$SME_EMB" \

#   --in_tar  "$NOB_EMB" \

#   --src_lid sme \

#   --tar_lid nob \

#   --idstring "EXP_sme_standard_nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_score" \

#   --out_src "${ALIGN_DIR}/aligned_sme_standard.vec" \

#   --out_tar "${ALIGN_DIR}/aligned_nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}.vec" \

#   --model_filename "${MODEL_DIR}/classifier_model_sme_standard_nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_score.pkl" \

#   --use_mnns_pooler 1 





# # ClassyMap evaluation (Single Dict)

# srun python3 ClassyMap/src/eval.py \

#   "${ALIGN_DIR}/aligned_sme_standard.vec" \

#   "${ALIGN_DIR}/aligned_nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}.vec" \

#   --dictionary "$TEST_DICT_SINGLE" \

#   --output_file "${RESULTS_DIR}/single_candidates__sme_standard_nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_score.txt" \

#   --src_lid sme \

#   --tar_lid nob \

#   --idstring "EXP_sme_standard_nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_score" \

#   --retrieval csls \

#   --seed 42 \

#   --super \

#   --cuda \

#   --model "${MODEL_DIR}/classifier_model_sme_standard_nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_score.pkl" \



# # ClassyMap evaluation (Multi Dict)

# srun python3 ClassyMap/src/eval.py \

#   "${ALIGN_DIR}/aligned_sme_standard.vec" \

#   "${ALIGN_DIR}/aligned_nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}.vec" \

#   --dictionary "$TEST_DICT_MULTI" \

#   --output_file "${RESULTS_DIR}/multi_candidates_sme_standard_nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_score.txt" \

#   --src_lid sme \

#   --tar_lid nob \

#   --idstring "EXP_sme_standard_nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_score" \

#   --retrieval csls \

#   --seed 42 \

#   --cuda \

#   --super \

#   --model "${MODEL_DIR}/classifier_model_sme_standard_nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_score.pkl" \



# echo "Finished: SME = Standard, NOB = minCount=$mc | minn=$mn | maxn=$mx"



# echo "Konfuzius sagt: \"Wo Hyperparameter variieren, wächst der Baum der Weisheit.\""







# echo "---------------------------------------------"

# echo "testing SME = minCount=$mc | minn=$mn | maxn=$mx, NOB = Standard"

# echo "---------------------------------------------"



# SME_EMB="${EMB_DIR}/sme_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}.vec"

# NOB_EMB=$NOB_EMB_STANDARD



# # ClassyMap training

# srun python3 ClassyMap/src/classymap.py \

#   --train_dict "$SEED_DICT" \

#   --in_src  "$SME_EMB" \

#   --in_tar  "$NOB_EMB" \

#   --src_lid sme \

#   --tar_lid nob \

#   --idstring "EXP_sme_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_nob_standard" \

#   --out_src "${ALIGN_DIR}/aligned_sme_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}.vec" \

#   --out_tar "${ALIGN_DIR}/aligned_nob_standard" \

#   --model_filename "${MODEL_DIR}/classifier_model_sme_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_nob_standard.pkl" \

#   --use_mnns_pooler 1 \



# # ClassyMap evaluation (Single Dict)

# srun python3 ClassyMap/src/eval.py \

#   "${ALIGN_DIR}/aligned_sme_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}.vec" \

#   "${ALIGN_DIR}/aligned_nob_standard" \

#   --dictionary "$TEST_DICT_SINGLE" \

#   --output_file "${RESULTS_DIR}/single_candidates_sme_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_nob_standard_test.txt" \

#   --src_lid sme \

#   --tar_lid nob \

#   --idstring "EXP_sme_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_nob_standard" \

#   --retrieval csls \

#   --seed 42 \

#   --cuda \

#   --super \

#   --model "${MODEL_DIR}/classifier_model_sme_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_nob_standard.pkl" \



# # ClassyMap evaluation (Multi Dict)

# srun python3 ClassyMap/src/eval.py \

#   "${ALIGN_DIR}/aligned_sme_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}.vec" \

#   "${ALIGN_DIR}/aligned_nob_standard" \

#   --dictionary "$TEST_DICT_MULTI" \

#   --output_file "${RESULTS_DIR}/multi_candidates_sme_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_nob_standard_test.txt" \

#   --src_lid sme \

#   --tar_lid nob \

#   --idstring "EXP_sme_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_nob_standard" \

#   --retrieval csls \

#   --seed 42 \

#   --cuda \

#   --super \

#   --model "${MODEL_DIR}/classifier_model_sme_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}_nob_standard.pkl" \



# echo "Finished: SME = minCount=$mc | minn=$mn | maxn=$mx, NOB = Standard"



# echo "Konfuzius sagt: \"Die GPU mag mächtig sein, doch auch sie folgt den Gesetzen der Geduld.\""







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

echo "testing SME = minCount=$mcs | minn=$mns | maxn=$mxs, NOB = minCount=$mc | minn=$mn | maxn=$mx"

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



SME_EMB=$SME_EMB_STANDARD

#"${EMB_DIR}/sme_dim${DIM}_mc${mcs}_minn${mns}_maxn${mxs}.vec"

NOB_EMB=$NOB_EMB_STANDARD

#"${EMB_DIR}/nob_dim${DIM}_mc${mc}_minn${mn}_maxn${mx}.vec"



# ClassyMap training

srun python3 ClassyMap/src/classymap.py \

  --train_dict "$SEED_DICT" \

  --in_src  "$SME_EMB" \

  --in_tar  "$NOB_EMB" \

  --src_lid sme \

  --tar_lid nob \

  --idstring "EXP_combined_all_classy" \

  --out_src "${ALIGN_DIR}/aligned_sme_combined_all_classy.vec" \

  --out_tar "${ALIGN_DIR}/aligned_nob_combined_all_classy.vec" \

  --model_filename "${MODEL_DIR}/classifier_model_combined_all_classy.pkl" \

  --use_mnns_pooler 1



# ClassyMap evaluation (Single Dict)

srun python3 ClassyMap/src/eval.py \

  "${ALIGN_DIR}/aligned_sme_combined_all_classy.vec" \

  "${ALIGN_DIR}/aligned_nob_combined_all_classy.vec" \

  --dictionary "$TEST_DICT_SINGLE" \

  --output_file "${RESULTS_DIR}/single_candidates_combined_all_classy.txt" \

  --src_lid sme \

  --tar_lid nob \

  --idstring "EXP_combined_all_classy" \

  --retrieval csls \

  --seed 42 \

  --cuda \

  --super \

  --model "${MODEL_DIR}/classifier_model_combined_all_classy.pkl" \



# ClassyMap evaluation (Multi Dict)

srun python3 ClassyMap/src/eval.py \

  "${ALIGN_DIR}/aligned_sme_combined_all_classy.vec" \

  "${ALIGN_DIR}/aligned_nob_combined_all_classy.vec" \

  --dictionary "$TEST_DICT_MULTI" \

  --output_file "${RESULTS_DIR}/multi_candidates_combined_all_classy.txt" \

  --src_lid sme \

  --tar_lid nob \

  --idstring "EXP_combined_all_classy" \

  --retrieval csls \

  --seed 42 \

  --cuda \

  --super \

  --model "${MODEL_DIR}/classifier_model_combined_all_classy.pkl" \



echo "Finished: SME = minCount=$mcs | minn=$mns | maxn=$mxs, NOB = minCount=$mc | minn=$mn | maxn=$mx"



echo "Konfuzius sagt: \"Die GPU mag mächtig sein, doch auch sie folgt den Gesetzen der Geduld.\""