Command_line_interface.md

Command line interface

usage: run_exp.py [-h] [-c CFG_FILE] [--weight WEIGHT] [--test_only] [--lr LR] [--bs BS] [--resume RESUME]
              [-m MAX_EPOCH] [--set ...]

deep learning on personality

optional arguments:
    -h, --help            show this help message and exit
    -c , --cfg_file       experiment config file
    --resume              saved model path to last training epoch
    --test_only           only test model on specified weights
    --weight              initialize with pretrained model weights
    --lr                  learning rate
    --bs                  training batch size
    -m, --max_epoch       set max training epochs
    --set ...             set config keys

Training sample

If we want to start an experiment, training can be triggered by corresponding config file

# <DeepPersonality as the top dir>
script/run_exp.py \
--cfg_file config/unified_frame_images/03_bimodal_resnet18.yaml 

Resume sample

If we want to resume training from a certain training checkpoint(saved model weights), parameter resume can be specified along with the saved weights. And before re-training, the training epochs and learning rate can be reset again if needed.

# <DeepPersonality as the top dir>
script/run_exp.py \
-c config/unified_frame_images/03_bimodal_resnet18.yaml \
--resume results/unified_frame_images/03_bimodal_resnet/12-19_18-15/checkpoint_199.pkl \
--max_epoch 210 \
--lr 0.001

Test sample

If we only want to test a trained model, parameter test_only can be used, and along with set parameters to specify the model weights used, shown as below:

script/run_exp.py \
-c config/unified_frame_images/09_hrnet.yaml \
--test_only \
--set TEST.WEIGHT results/unified_frame_images/09_hrnet/12-20_22-12/checkpoint_186.pkl