Update Stage1 script and Add Stage2 script

script_stage1/run_mimic_abn.sh

@@ -23,7 +23,7 @@ beta=1
 dataloader_num_workers=8
 log_level="info"
 report_to="wandb"
-chexbert_label="/CheXbert/mimic_abn/id2tag.csv"
+chexbert_label="./CheXbert/mimic_abn/id2tag.csv"
 annotation_file="./mimic_abn/annotation.json"
 miss_annotation_file="./mimic_abn/miss_annotation.json"
 output_dir="./tmp_stage1/mimic_abn_stage1/"

script_stage2/run_mimic_abn.sh

@@ -0,0 +1,115 @@
+#!/bin/sh
+export CUDA_VISIBLE_DEVICES=0
+export TOKENIZERS_PARALLELISM=true
+export TRANSFORMERS_OFFLINE=false
+suffix=""
+warmup_ratio=0.0
+max_tgt_length=64
+num_train_epochs=10
+overwrite_output_dir=false
+evaluation_strategy=epoch
+per_device_train_batch_size=32
+per_device_eval_batch_size=32
+gradient_accumulation_steps=1
+debug_model=false
+seed=42
+num_beams=4
+slow_lr=5e-5
+fast_lr=1e-4
+weight_decay=0.0
+alpha=2
+beta=1
+dataloader_num_workers=8
+log_level="info"
+report_to="none"
+chexbert_label="./CheXbert/mimic_abn/id2tag.csv"
+annotation_file="./mimic_abn/annotation.json"
+miss_annotation_file="./mimic_abn/miss_annotation.json"
+stage1_model_name_or_path=$2
+stage1_eval_file=$3
+graph_version=$4
+progression_graph="./data/${graph_version}/mimic_abn/triples.json"
+date=$5
+topk=$6
+lambda_=$7
+output_dir="./tmp_stage2/mimic_abn_ablation_${date}_top${topk}_lambda${lambda_}/"
+
+
+
+if [ "$wo_obs" -ne 1 ];
+then
+    echo "********** Load Stage 1 Model **********"
+    echo "********** Load Stage 1 Model **********"
+    echo "********** Load Stage 1 Model **********"
+    num_train_epochs=5
+fi
+
+if [ "$1" -ne 1 ];
+then
+    echo "********** debug **********"
+    echo "********** debug **********"
+    echo "********** debug **********"
+    suffix="_debug"
+    num_train_epochs=1
+    output_dir="./tmp/bert_doc_baseline_debug"
+    overwrite_output_dir=true
+    debug_model=true
+    report_to="none"
+fi
+
+export TOKENIZERS_PARALLELISM=true
+python3 -u ./src_stage2/run_ende.py \
+    --chexbert_model_name_or_path ./CheXbert/chexbert.pth \
+    --stage1_model_name_or_path $stage1_model_name_or_path \
+    --stage1_eval_file $stage1_eval_file \
+    --test_model_name_or_path $output_dir \
+    --annotation_file $annotation_file \
+    --miss_annotation_file $miss_annotation_file \
+    --graph_version $graph_version \
+    --progression_graph $progression_graph \
+    --history "./mimic_abn/temporal_ids.json" \
+    --image_path ./mimic_cxr/images/ \
+    --chexbert_label $chexbert_label \
+    --is_stage1_pretrained 0 \
+    --is_temporal 1 \
+    --wo_op $wo_op \
+    --wo_obs $wo_obs \
+    --wo_pro $wo_pro \
+    --wo_prr $wo_prr \
+    --topk $topk \
+    --lambda_ $lambda_ \
+    --do_train \
+    --do_eval \
+    --do_predict \
+    --log_level $log_level \
+    --per_device_train_batch_size $per_device_train_batch_size \
+    --gradient_accumulation_steps $gradient_accumulation_steps \
+    --per_device_eval_batch_size $per_device_eval_batch_size \
+    --max_tgt_length $max_tgt_length \
+    --output_dir $output_dir \
+    --warmup_ratio $warmup_ratio \
+    --num_train_epochs $num_train_epochs \
+    --learning_rate $slow_lr \
+    --fast_lr $fast_lr \
+    --weight_decay $weight_decay \
+    --evaluation_strategy $evaluation_strategy \
+    --save_strategy $evaluation_strategy \
+    --save_total_limit 1 \
+    --alpha $alpha \
+    --beta $beta \
+    --seed $seed \
+    --logging_steps 100 \
+    --report_to $report_to \
+    --fp16 \
+    --fp16_opt_level O2 \
+    --fp16_full_eval \
+    --dataloader_num_workers $dataloader_num_workers \
+    --load_best_model_at_end true \
+    --overwrite_output_dir $overwrite_output_dir \
+    --group_by_length false \
+    --length_column_name length \
+    --eval_on_gen \
+    --greater_is_better true \
+    --metric_for_best_model eval_BLEU_4 \
+    --debug_model $debug_model \
+    --num_beams $num_beams