update gen_meta_data.sh

scripts/gen_meta_data.sh

@@ -3,16 +3,16 @@
 echo usage: pass dataset list as param, split with space
 echo eg: source gen_mate_data.sh atis
 
-dataset_lst=(smp)
+dataset_lst=(snips)
 #dataset_lst=(toursg)
 #dataset_lst=(stanford)
 #dataset_lst=(atis stanford toursg)
 
 # ======= size setting ======
-support_shots_lst=(1)
-#support_shots_lst=(5)
-episode_num=50  # We could over generation and select part of for each epoch
-query_shot=4
+#support_shots_lst=(1)
+support_shots_lst=(5)
+episode_num=100  # We could over generation and select part of for each epoch
+query_shot=20
 word_piece_data=True
 way=-1
 
@@ -34,9 +34,9 @@ allow_override=--allow_override
 check=--check
 
 # ====== train & test setting ======
-split_basis=domain
+split_basis=sent_label
 
-#eval_confif_id_lst=(1)  # for snips
+eval_config_id_lst=(0 1 2 3 4 5 6)  # for snips
 #eval_config_id_lst=(0 1 2 3 4 5)  # for toursg
 label_type_lst=(attribute)
 
@@ -46,8 +46,9 @@ use_fix_support=--use_fix_support
 # ======= default path (for quick distribution) ==========
 #input_dir=/Users/lyk/Work/Dialogue/FewShot/SMP正式数据集/
 #output_dir=/Users/lyk/Work/Dialogue/FewShot/SMP正式数据集/SmpMetaData/
-input_dir=/Users/lyk/Work/Dialogue/FewShot/SMP/SMP_Final_Origin2
-output_dir=/Users/lyk/Work/Dialogue/FewShot/SMP/smp_final/
+input_dir=/Users/niloofarasadi/PycharmProjects/MetaDialog/data/original/snips/
+output_dir=/Users/niloofarasadi/PycharmProjects/MetaDialog/data/k-shot/
+config_dir=/Users/niloofarasadi/PycharmProjects/MetaDialog/data/config/
 
 now=$(date +%s)
 
@@ -59,30 +60,30 @@ do
   do
     for support_shots in ${support_shots_lst[@]}
     do
+      for eval_config_id in ${eval_config_id_lst[@]}
+      do
       echo \[CLI\] generate with \[ ${use_fix_support} \]
       input_path=${input_dir}
       mark=try
 
-      export OMP_NUM_THREADS=2  # threads num for each task
-      python3 ./other_tool/meta_dataset_generator/generate_meta_dataset.py \
-        --input_path ${input_path} \
-        --output_dir ${output_dir} \
-        --dataset ${dataset} \
-        --episode_num ${episode_num} \
-        --support_shots ${support_shots} \
-        --query_shot ${query_shot} \
-        --way ${way} \
-        --task ${task} \
-        --seed ${seed} \
-        --split_basis ${split_basis} \
-        --remove_rate ${remove_rate} \
-        ${use_fix_support} \
-        --mark ${mark} ${dup_query} ${allow_override} ${check} > ${output_dir}${dataset}.spt_s_${support_shots}.q_s_${query_shot}.ep_${episode_num}${use_fix_support}-${now}.log
+        export OMP_NUM_THREADS=2  # threads num for each task
+        python3 ./other_tool/meta_dataset_generator/generate_meta_datase.py \
+          --input_path ${input_path} \
+          --output_dir ${output_dir} \
+          --dataset ${dataset} \
+          --episode_num ${episode_num} \
+          --support_shots ${support_shots} \
+          --query_shot ${query_shot} \
+          --way ${way} \
+          --task ${task} \
+          --seed ${seed} \
+          --split_basis ${split_basis} \
+          --remove_rate ${remove_rate} \
+          ${use_fix_support} \
+          --eval_config ${config_dir}config${eval_config_id}.json\
+          --eval_config_id ${eval_config_id}\
+          --mark ${mark} ${dup_query} ${allow_override} ${check} > ${output_dir}${dataset}.spt_s_${support_shots}.q_s_${query_shot}.ep_${episode_num}${use_fix_support}-${now}.log
+      done
     done
   done
 done
-
-
-
-
-

scripts/other_tool/meta_dataset_generator/generate_meta_dataset.py

@@ -98,13 +98,13 @@ def split_eval_set_with_label(opt, raw_data):
     """ Both input and output are raw data format """
     try:
         # load config:
-        with open(opt.eval_labels, 'r') as reader:
+        with open(opt.eval_config, 'r') as reader:
             config = json.load(reader)
         # split labels
         train_data = {'seq_ins': [], 'labels': [], 'seq_outs': []}
         dev_data = {'seq_ins': [], 'labels': [], 'seq_outs': []}
         test_data = {'seq_ins': [], 'labels': [], 'seq_outs': []}
-        for old_domain_name, old_domain in raw_data:
+        for old_domain_name, old_domain in raw_data.items():
             for ind in range(len(old_domain['label'])):
                 seq_ins, seq_outs, labels = old_domain['seq_ins'][ind], old_domain['seq_outs'][ind], old_domain['labels'][ind]
                 # for label in labels: