CUDA_VISIBLE_DEVICES=0 python main.py \
--mode validate \
--config-yml $CONFIG \
--checkpoint $CKPT_PATH \
--validate-data $DATA_SRC \
--validate-batch-size 16 \
--separate-evaluation \
--shuffle \
You should specify CONFIG, CKPT_PATH, and DATA_SRC.
The arguments mean as follows:
--config-yml: a configuration file that was used to train the checkpoint (which was automatically saved at the checkpoints directory of this training result)--checkpoint: a checkpoint of an MLFF model--validate-data: a data source path, which can be a single .lmdb file or a directory including multiple .lmdb files--validate-batch-size N(optional) : a batch size of N used for validation (if not given, it is set as a batch size in the configuration file)--separate-evaluation(optional) : print the results of multiple data files separately (default: False)--shuffle(optional) : shuffle the data (default: False)
./run_validate.sh $GPU $CKPT_DIR $DATA_LMDB
# example:
# ./run_validate.sh 0 ../../train_results/SiN/GemNet-dT/checkpoints/train-20230101_010000 ../../datasets/SiN/ood/atom_cloud/ood.lmdb
In this script, the checkpoint used for validation is the last checkpoint.