This is a variation of the Liang model found in https://github.com/yuliang419/Astronet-Vetting.
Generate new input data:
python astronet/data/generate_input_records.py --input_tce_csv_file=astronet/tces-new+old.
csv --tess_data_dir=/home/${USER}/lc --output_dir=astronet/tfrecords-new+old --num_worker_processes=8 --make_test_set
Train ensemble (modify the .sh file to match the output_dir above):
./astronet/ensemble_train.sh
Tune (requires some setup, see Tune.ipynb):
python astronet/tune.py --model=AstroCNNModel --config_name=local_global_new --train_files=astronet/tfrecords-new\+old/test-0000[0-5]* --eval_files=astronet/tfrecords-new\+old/test-0000[6-6]* --train_steps=7000 --tune_trials=1000 --client_secrets=${HOME}/client_secrets.json --study_id=a_unique_string_id
Run predictions (or use Predict.ipynb for one-offs):
python astronet/predict.py --model_dir=/tmp/astronet/AstroCNNModel_local_global_multiclass_20200222_154634 --data_files=astronet/tfrecords-new\+old/* --output_file=/home/${USER}/predictions.csv