Check the SONYC_UST_PATH in "download_data.sh" and "train_start.sh"

run ./dwonload_data.sh

run ./train_start.sh

pip3 install -r requirements.txt

Data Prepare shell

./download_data.sh

Main Sequence shell

./train_start.sh

Prepare the feature for train

python3 src/feature_extraction.py [audio_folder_path] [feature_output_dir]

Train, create Model and predict validate set

python3 src/classify.py [annotation_path] [taxonomy_path] [output_dir] [--emb_dir] [--ef_mode] [--dropout_size] [--learning_rate] [--l2_reg] [--batch_size] [--num_epochs] [--patience] [--random-state]

Evaluate the predicted validate set

python3 src/evaluate_predictions.py [output_dir] [annotation_path] [taxonomy_path]

Best result of task5 in our model

Fine : best_evaluation_fine.json

Coarse : best_evaluation_coarse.json