ASiCo1116/ADL21-HW1

Homework 1 ADL NTU 110-2 Spring

Environment

# If you have pipenv, you can build the environemt from pipfile
pipenv install --skip-lock

# Otherwise
pip install -r requirements.txt

Preprocessing

# To preprocess intent detectiona and slot tagging datasets
bash preprocess.sh

Intent classification

Training

# intent.sh
python3 train_intent.py \
    --data_dir ROOT_TO_INTENT_DATA \
    --cache_dir ROOT_TO_INTENT_CACHE \
    --ckpt_dir ROOT_TO_SAVE_CHECKPOINT\
    --lr LR \
    --batch_size BS \
    --hidden_size HS \
    --num_layers NL \
    --num_epoch NE \
    --device DEVICE

Testing

# intent_cls.sh
python3.8 test_intent.py --test_file ROOT_TO_TEST_FILE \
    --cache_dir ROOT_TO_INTENT_CACHE \
    --ckpt_path ROOT_TO_CHECKPOINT \
    --pred_file ROOT_TO_SAVE_PREDICTION \
    --batch_size BS \
    --hidden_size HS \
    --num_layers NL \
    --device DEVICE

Slot tagging

Training

# slot.sh
python3 train_slot.py \
    --data_dir ROOT_TO_SLOT_DATA \
    --cache_dir ROOT_TO_SLOT_CACHE \
    --ckpt_dir ROOT_TO_SAVE_CHECKPOINT\
    --lr LR \
    --batch_size BS \
    --hidden_size HS \
    --num_layers NL \
    --num_epoch NE \
    --max_len ML \
    --device DEVICE

Testing

# slot_tag.sh
python3.8 test_slot.py --test_file ROOT_TO_TEST_FILE \
    --cache_dir ROOT_TO_SLOT_CACHE \
    --ckpt_path ROOT_TO_CHECKPOINT \
    --pred_file ROOT_TO_SAVE_PREDICTION \
    --batch_size BS \
    --hidden_size HS \
    --max_len ML \
    --num_layers NL \
    --device DEVICE