TIME_LIMIT=${1}

ITERATIONS=${2}

OUTPUT_DIR=${3}

PYTHON_COMMAND=${4}

declare -a models=("linear" "nn16" "nn32" "nn64" "nn128")

WORKDIR="."

echo "Start training data on heuristical based selectors"

$WORKDIR/scripts/train_data.sh $TIME_LIMIT

echo "Start iterative learning of models"

for (( i=0; i < $ITERATIONS; i++ ))

echo "Start $i iteration"

for model in "${models[@]}"

do

EXTRA_ARGS=""

if [[ $model == *"nn"* ]]; then

EXTRA_ARGS="--hidden_dim $(echo $model | cut -c 3-)"

echo "EXTRA_ARGS=$EXTRA_ARGS"

fi

COMMAND="$PYTHON_COMMAND $WORKDIR/scripts/train.py --features_dir $WORKDIR/eval/features --output_dir $OUTPUT_DIR/$model/$i --prog_list $WORKDIR/scripts/prog_list --model $model $EXTRA_ARGS"

echo "TRAINING COMMAND=$COMMAND"

$COMMAND

done

while read prog; do

prog="${prog%%[[:cntrl:]]}"

for model in "${models[@]}"

do

PREDICTOR="BASE"

if [[ $model == *"linear"* ]]; then

PREDICTOR="LINEAR"

fi

$WORKDIR/scripts/run_contest_estimator.sh $prog $TIME_LIMIT "NN_REWARD_GUIDED_SELECTOR $OUTPUT_DIR/$model/$i $PREDICTOR" "true eval/features/jlearch/$model$i/$prog"

done

done <"$WORKDIR/scripts/prog_list"