EXPERIMENTS.md

Experiments

This document keeps track of experiment runs during the course of this project.

• Debugging Experiment: Is the training loss going down?

  • Run Name: merciful-frog-774
  • Date: 05/26/2021
  • Reproduce:

• Debugging Experiment: Overfitting on a small batch of data:

  • Run Names:
    • nostalgic-impala-272
    • upbeat-iguana-246
    • cornflower-rooster-462
    • mellow-gibbon-813
    • economic-tuatara-996
  • Date: 6/12/2021 - 8/7/2021
  • Reproduce:
    • 
    • 
  • 
  • [](https://platform.grid.ai/#/runs?script=https://github.com/cosmicBboy/movenet/blob/512ec692bfeb017f367c2d94ae83f77681b889dd/movenet/trainer.py&cloud=grid&instance=t2.2xlarge&accelerators=7&framework=torch&script_args=movenet/trainer.py --dataset /opt/datastore --n_training_steps 500 --learning_rate 0.0003 --input_channels 64 --residual_channels 64 --layer_size 3 --stack_size 3 --checkpoint_every 25 --model_output_path models --pretrained_model_path /artifacts/mellow-gibbon-813-exp0/models/model.pth --pretrained_run_exp_name mellow-gibbon-813-exp0&datastore_name=kinetics-debug&datastore_version=5&datastore_id=cda1f319-b794-48a7-99b5-6804b6d00440&datastore_creator=cosmicBboy&datastore_mount_dir=/opt/datastore)

• Learning Rate Range Experiment:

  • https://wandb.ai/nielsbantilan/dance2music/runs/3a4sfxev?workspace=user-nielsbantilan
  • grid.ai run: complex-roentgen-3889
  • grid.ai experiment: complex-roentgen-3889-exp0