# ================== Solver list ================== #

# Our Algorithms: [flag_vne, flag_vne_meta_free_single_policy, flag_vne_meta_free_multi_policy, flag_vne_meta_policy, flag_vne_no_curriculum]

# ================== 1. Key Settings ================== #

solver_name="flag_vne" # Solver name. Options: SOLVER_LIST

topology="geant" # Topology name. Options: [geant, wx100]

num_train_epochs=5000 # Number of training epochs. Options: [0, >0]. If 0, then inference only.

num_meta_learning_epochs=0 # Number of meta learning epochs. Only work for FlagVNE and its variants.

# ================== 2. Simulation Settings ================== #

v_sim_setting_num_v_nets=1

unseen_v_net_size=12

v_sim_setting_v_net_size_low=$unseen_v_net_size

v_sim_setting_v_net_size_high=$unseen_v_net_size

use_pretrained_model=1

# ================ 3. Other Settings ================ #

cuda_device=0 # Cuda device id

batch_size=128 # Batch size

# ===================================================== #

identifier="-train_epochs_$num_train_epochs-v_sim_setting_num_v_nets_$v_sim_setting_num_v_nets-meta_learning_epochs_$num_meta_learning_epochs-$solver_pretrained_model_name-$use_pretrained_model"

# set pretrained model path for testing

declare -A pretrained_model_path_dict_for_geant

declare -A pretrained_model_path_dict_for_wx100

pretrained_model_path_dict_for_geant["SOLVE_NAME"]="PERTRAINED_MODEL_PATH"

pretrained_model_path_dict_for_wx100["SOLVE_NAME"]="PERTRAINED_MODEL_PATH"

if [ $topology == "geant" -a $use_pretrained_model == 1 ]; then

pretrained_model_path=${pretrained_model_path_dict_for_geant[$solver_name]}

elif [ $topology == "wx100" -a $use_pretrained_model == 1 ]; then

pretrained_model_path=${pretrained_model_path_dict_for_wx100[$solver_name]}

else

pretrained_model_path="null"

fi

echo $pretrained_model_path

if [ $topology == "geant" ]; then

# geant topology

if [ $num_train_epochs == "0" ]; then

# inference setting

aver_arrival_rate_list=$(seq 0.001 0.001 0.012)

identifier="-test-$solver_pretrained_model_name-$use_pretrained_model$identifier"

else

# pretrain setting

aver_arrival_rate_list=(0.001)

# pretrained_model_path="null"

fi

elif [ $topology == "wx100" ]; then

# wx100 topology

if [ $num_train_epochs == "0" ]; then

# inference setting

aver_arrival_rate_list=$(seq 0.08 0.02 0.18)

identifier="-test-$solver_pretrained_model_name-$use_pretrained_model$identifier"

echo $pretrained_model_path

else

# pretrain setting

aver_arrival_rate_list=(0.08)

# pretrained_model_path="null"

fi

else

echo "Error: topology $topology is not supported!"

exit 1

fi

# Judge if the pretrained model exists. If inference, then the path must be valid.

if [ "$pretrained_model_path" == "null" ]; then

echo "pretrained model path is null, skip the check"

else

if [ ! -f $pretrained_model_path ]; then

echo "Error: pretrained model $pretrained_model_path does not exist!"

exit 1

fi

fi

for aver_arrival_rate in $aver_arrival_rate_list

do

echo "aver_arrival_rate: $aver_arrival_rate"

CUDA_VISIBLE_DEVICES=$cuda_device \

python main.py \

--p_net_topology=$topology \

--solver_name=$solver_name \

--num_train_epochs=$num_train_epochs \

--num_meta_learning_epochs=$num_meta_learning_epochs \

--eval_interval=$num_train_epochs \

--save_interval=$num_train_epochs \

--v_sim_setting_num_v_nets=$v_sim_setting_num_v_nets \

--v_sim_setting_v_net_size_low=$v_sim_setting_v_net_size_low \

--v_sim_setting_v_net_size_high=$v_sim_setting_v_net_size_high \

--pretrained_model_path=$pretrained_model_path \

--batch_size=$batch_size \

--v_sim_setting_aver_arrival_rate=$aver_arrival_rate \

--verbose=1 \

--lr_actor=0.001 \

--lr_critic=0.001 \

--summary_dir="exp_data/flag_vne/adaptation_12" \

--summary_file_name="$topology-$solver_name$identifier.csv"

done