README.md

Code Organization

All code is under src/seqmodel. All unit tests are under test/seqmodel. Data files included with tests are in test/data.

• data contains genomic data.
• src/seqmodel contains code.
  • seqmodel.experiment contains individual experiments and hyperparameters
  • seqmodel.functional contains functions for preprocessing data for specific tasks
    • seqmodel.functional.log contains functions for summarizing tensors and outputting loggable strings
    • seqmodel.functional.transform contains common sequence transformations which are composable
  • seqmodel.model contains model components based on nn.Module
  • seqmodel.seqdata contains functions and objects for loading data
• test/data contains data files for tests.
• test/seqmodel contains unit tests.

Installation

Requires python 3.6> and cuda/10.1. Uses pytorch==1.5. Install dependencies using:

pip install --upgrade pip
pip install -r $SOURCE_DIR/requirements.txt

Git Guidelines

Developing

• Create and checkout a new branch:

  git branch feat/[development|experimental]/{feature_name}
  git checkout {branch_name}
  

• Develop features in src/** and unit tests in test/**.
• Commit cohesive units of code in progress:

  git add -A
  git commit
  

  Refer to git log for examples of previous commits. Follow these guidelines when writing commit messages:

  {tag [ADD|FIX|DOC|STYLE|REFACTOR|TEST|CHORE]} {summary no more than 50 characters}
  {empty line}
  {tag} {paragraph explaining detail}
  {empty line}
  (optional {tag [TODO|FIXME]} {additional paragraphs for other details. Use TODO to indicate WIP and FIXME to indicate issues.}
  {empty line})
  

• Make sure unit tests pass:

  python -m unittest discover
  

  Test integration by running on HPC.
• Once tests pass, merge and fix merge issues:

  git checkout main
  git merge {branch_name}
  

Running on HPC

• On HPC, create bare repository:

  mkdir {target_dir} {target_dir}/.git
  cd {target_dir}/.git
  git init --bare
  

• On HPC, add post-receive hook to checkout code after pushing.

  cat >> hooks/post-receivetest << EOL
  #!/bin/sh
  GIT_WORK_TREE={target_dir} git checkout -f
  
  EOL
  chmod +x hooks/post-receive
  

• On local machine, add remote:

  git remote add {remote_name} {username}@{HPC_url}:{target_dir}/.git
  

• To run a particular branch on HPC, checkout and push from the local machine:

  git checkout {branch_name}
  git push {remote_name} {branch_name}
  

Pytorch-Lightning

Default Command Line Arguments

These are for pytorch-lightning only, see individual model files (e.g. src/experiment/seqbert.py) for model-specific arguments.

• accumulate_grad_batches=1
• amp_backend='native'
• amp_level='O2'
• auto_lr_find=False
• auto_scale_batch_size=False
• auto_select_gpus=False
• benchmark=False
• check_val_every_n_epoch=1
• checkpoint_callback=True
• default_root_dir=None
• deterministic=False
• distributed_backend=None
• early_stop_callback=False
• fast_dev_run=False
• gpus=1
• gradient_clip_val=0
• limit_test_batches=1.0
• limit_train_batches=1.0
• limit_val_batches=1.0
• log_gpu_memory=None
• log_save_interval=100
• logger=True
• max_epochs=1000
• max_steps=None
• min_epochs=1
• min_steps=None
• num_nodes=1
• num_processes=1
• num_sanity_val_steps=2
• overfit_batches=0.0
• overfit_pct=None
• precision=32
• prepare_data_per_node=True
• process_position=0
• profiler=None
• progress_bar_refresh_rate=1
• reload_dataloaders_every_epoch=False
• replace_sampler_ddp=True
• resume_from_checkpoint=None
• row_log_interval=50
• sync_batchnorm=False
• terminate_on_nan=False
• test_percent_check=None
• tpu_cores=<function Trainer._gpus_arg_default at 0x7f614e8940d0>
• track_grad_norm=-1
• train_percent_check=None
• truncated_bptt_steps=None
• val_check_interval=1.0
• val_percent_check=None
• weights_save_path=None
• weights_summary='top'