diff --git a/.gitignore b/.gitignore index 63809da..d009932 100644 --- a/.gitignore +++ b/.gitignore @@ -38,6 +38,5 @@ pytest-*.xml *.log *.txt *.pyz -*.png *.metadata *.json diff --git a/AGENTS.md b/AGENTS.md new file mode 100644 index 0000000..57b0fb3 --- /dev/null +++ b/AGENTS.md @@ -0,0 +1,253 @@ +# AGENTS.md + +## Project overview + +PepSeqPred is a residue level epitope prediction pipeline for peptide and protein workflows. + +This repository is organized around: +- `src/pepseqpred/apps/` for user facing CLIs +- `src/pepseqpred/core/` for reusable pipeline logic +- `scripts/hpc/` for SLURM batch execution on GPU clusters +- `tests/` for `unit`, `integration`, and `e2e` coverage +- `envs/`, `localdata/`, `notebooks/`, and `dist/` as supporting project directories + +Primary goals when working in this repo: +- preserve scientific reproducibility +- keep training and evaluation behavior stable unless explicitly asked to change it +- prefer minimal, targeted edits +- avoid expensive or risky compute by default +- maintain compatibility with existing CLIs, scripts, and downstream outputs + +## Repository structure + +### Application entrypoints + +The main CLIs are: +- `pepseqpred-esm` +- `pepseqpred-labels` +- `pepseqpred-predict` +- `pepseqpred-preprocess` +- `pepseqpred-train-ffnn` +- `pepseqpred-train-ffnn-optuna` + +These map to files in `src/pepseqpred/apps/`. + +### Core package layout + +Important subpackages under `src/pepseqpred/core/`: +- `data/` for dataset loading +- `embeddings/` for ESM based embedding generation +- `io/` for logging and file writing helpers +- `labels/` for label generation logic +- `models/` for model definitions +- `predict/` for inference +- `preprocess/` for preprocessing workflows +- `train/` for DDP, splitting, metrics, thresholds, trainer logic, seeds, and class weighting + +### HPC scripts + +Batch scripts live in `scripts/hpc/`. These are part of the intended workflow, especially for: +- embedding generation +- label generation +- preprocessing +- prediction +- FFNN training +- FFNN Optuna tuning + +Treat these scripts as first class project interfaces, not throwaway helpers. + +## General working rules for any agents (Codex, Claude Code, etc.) + +Before editing: +- inspect the relevant files first +- understand the existing CLI and core flow before proposing changes +- prefer the smallest possible diff +- do not rename modules, scripts, CLI flags, or output files unless the task requires it +- do not introduce dependencies unless clearly justified + +While editing: +- follow the existing package structure +- preserve current naming conventions and CLI semantics +- preserve public script behavior unless the user explicitly asks for a behavior change +- keep functions explicit and readable +- add or update docstrings when behavior changes +- avoid unrelated refactors or cosmetic churn + +After editing: +- run the smallest relevant validation first +- report exactly what changed +- note anything you could not validate + +## Reproducibility and experiment safety + +This is research code. Changes can silently invalidate experiments. + +Always preserve: +- deterministic seed handling +- train, validation, and test split semantics +- masking behavior for uncertain labels +- metric calculation behavior +- checkpoint and result artifact formats, unless explicitly changing schema +- per run and per trial traceability + +Do not: +- change default seeds casually +- change label meaning or preprocessing behavior without documenting it +- mix outputs from different experiments into ambiguous files +- overwrite prior results when a new output path is safer + +If a change affects training or evaluation, explicitly check for: +- data leakage +- split leakage +- rank specific side effects +- output collisions across repeated runs or trials + +## Distributed training and HPC guardrails + +PepSeqPred training is designed around multi-GPU DistributedDataParallel and SLURM-based execution. + +When touching training or Optuna code: +- assume jobs may run on at least 4 GPUs through SLURM +- be careful with `torch.distributed` collectives, barriers, and rank scoped logic +- ensure shared artifacts are only written by the correct rank +- avoid introducing deadlocks +- do not make changes that multiply compute cost unexpectedly +- preserve scheduler-friendly behavior + +Prefer: +- local dry runs +- tiny subsets +- reduced epoch smoke tests +- single rank validation where possible before full scale recommendations + +Do not assume: +- local laptop training is practical +- interactive GPU access exists +- paths outside repo root are portable unless already established by project scripts +- using `sbatch` locally will work, it will fail for local development + +## Data and artifact handling + +Never modify raw or source data in place. + +Prefer: +- writing derived outputs to new paths +- append safe logs and result files +- explicit artifact names that encode experiment identity + +Be careful with: +- checkpoint directories +- CSV summaries +- Optuna trial outputs +- per-rank logging +- temporary files on shared scratch storage + +If a schema or file format must change: +- make the change explicit +- update readers and writers together +- document the migration clearly + +## Validation expectations + +Use the repo’s configured tooling where practical. + +Default validation order: +1. `ruff check .` +2. targeted `pytest` invocation for affected tests +3. broader `pytest` if the change is cross-cutting +4. only then consider heavier runtime checks + +Important: +- do not run long HPC style training jobs unless explicitly asked +- do not present expensive end-to-end training as routine validation +- for training code, prefer smoke tests over full experiments + +If validation is incomplete: +- say what was not run +- say why +- identify the main remaining risks + +## Commands + +Common commands: +- install package: `pip install -e .` +- install dev tools: `pip install -e .[dev]` +- run tests: `pytest` +- lint: `ruff check .` +- format: `ruff format .` + +Available CLIs: +- `pepseqpred-esm` +- `pepseqpred-labels` +- `pepseqpred-predict` +- `pepseqpred-preprocess` +- `pepseqpred-train-ffnn` +- `pepseqpred-train-ffnn-optuna` + +## Testing guidance + +The repo has: +- `tests/unit/` +- `tests/integration/` +- `tests/e2e/` + +Prefer: +- unit tests for isolated logic changes +- integration tests for CLI to core interactions +- e2e only when a full pipeline boundary changed + +Do not expand test scope unnecessarily if a small targeted test is enough. + +## Documentation expectations + +When behavior changes, update the relevant: +- docstrings +- CLI help text +- comments near tricky distributed logic +- any usage examples affected by the change + +Note: +- the current root `README.md` is minimal, so do not assume broader user documentation already exists +- if you add a major new workflow, include enough inline guidance for future contributors + +## What not to change without explicit approval + +Do not, unless clearly requested: +- redesign package structure +- replace DDP or SLURM workflows +- alter default experiment semantics +- change model architecture defaults broadly +- change preprocessing formulas or label logic +- rewrite output schemas +- remove test categories +- introduce large framework migrations + +## Preferred task workflow + +For most tasks: +1. inspect relevant app, core, test, and script files +2. identify the smallest safe fix +3. implement minimally +4. run focused validation +5. summarize edits, validation, and remaining risks + +## Directory specific notes + +### `src/pepseqpred/apps/` +- preserve CLI compatibility +- do not break argument names or defaults without explicit instruction +- keep orchestration logic thin when possible + +### `src/pepseqpred/core/train/` +- highest risk area +- be conservative with splits, seeds, metrics, thresholds, and DDP behavior +- verify rank aware writes and collective calls carefully + +### `scripts/hpc/` +- preserve SLURM semantics +- avoid hard coding user specific assumptions unless already part of script conventions +- comment any scheduler related changes clearly + +### `tests/` +- add targeted coverage for bug fixes +- do not rewrite unrelated fixtures or tests just for style \ No newline at end of file diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md new file mode 100644 index 0000000..ae80fc6 --- /dev/null +++ b/CONTRIBUTING.md @@ -0,0 +1,101 @@ +# Contributing to PepSeqPred + +This document defines required contribution workflow, naming conventions, and pull request expectations for this repository. + +## Core Rules + +- Do not develop directly on `main`. +- All changes must be made on a different branch and merged via pull request. +- Branch, issue, and commit names must follow the conventions below. +- Keep titles and descriptions short, clear, and specific. + +## Required Contribution Workflow + +1. Create or confirm an issue for the work. +2. Create a branch from the latest `main` using the branch naming rules. +3. Implement the change and add/update relevant tests. +4. Run required local checks. +5. Open a pull request into `main` with required summary and verification details. + +## Branch Naming Conventions + +Use lowercase and hyphen-separated descriptions. + +Accepted patterns: +- `feat/short-description` +- `fix/short-description` +- `docs/short-description` +- `chore/short-description` +- `test/short-description` +- `refactor/short-description` + +You can also be extra specific by adding the issue number associated with your code as seen below. + +Examples: +- `feat/add-sharded-embedding-index-logging` +- `fix/issue-42-threshold-range-validation` +- `docs/update-readme-pipeline-section` + +## Issue Naming and Content + +Issue title format: +- `: short description` + +Examples: +- `bug: label shard mismatch across embedding keys` +- `docs: add hpc setup troubleshooting` +- `chore: tighten local test gating in README` + +Issue body requirements: +- `Summary`: a short statement of the problem or request. +- `Done when`: acceptance criteria, if applicable. + +## Commit Message Conventions + +Commit title format: +- `: short description` + +Examples: +- `bug: fix id-family key validation in labels builder` +- `chore: remove unused import from prediction cli` +- `docs: add contributing workflow and naming rules` + +Commit guidance: +- Keep the first line concise and specific. +- Keep one logical change per commit where possible. + +## Pull Request Requirements + +All pull requests to `main` must include: +- A concise summary of what changed. +- Linked issue(s) (for example, `Fixes #42`). +- A concise "How to verify" section with exact commands. +- Any new or updated unit, integration, or e2e tests needed to verify behavior changes. + +PRs should not include changed unrelated to the issue unless it's minor, please use your own discretion. + +## Verification Expectations Before PR + +Run these checks locally before opening a PR: + +```bash +ruff check . +pytest -m "unit or integration or e2e" +``` + +If behavior changed, include targeted test commands in the PR verification section, along with expected outcomes. + +## PR Checklist + +- [ ] Branch name follows convention. +- [ ] Issue title/body follow convention (`Summary` and `Done when` included when applicable). +- [ ] Commit messages follow `: short description`. +- [ ] No development occurred directly on `main`. +- [ ] PR includes concise summary and reproducible verification steps. +- [ ] Relevant unit/integration/e2e tests were added or updated. + +## Maintainer Support and Escalation + +- Use GitHub issues for normal development questions, bug reports, and feature requests. +- Use email for private or sensitive matters that should not be posted publicly. +- Maintainer contact: [Jeffrey Hoelzel](mailto:jmh2338@nau.edu) or [Jason Ladner](mailto:jason.ladner@nau.edu). diff --git a/LICENSE b/LICENSE new file mode 100644 index 0000000..f288702 --- /dev/null +++ b/LICENSE @@ -0,0 +1,674 @@ + GNU GENERAL PUBLIC LICENSE + Version 3, 29 June 2007 + + Copyright (C) 2007 Free Software Foundation, Inc. + Everyone is permitted to copy and distribute verbatim copies + of this license document, but changing it is not allowed. + + Preamble + + The GNU General Public License is a free, copyleft license for +software and other kinds of works. + + The licenses for most software and other practical works are designed +to take away your freedom to share and change the works. By contrast, +the GNU General Public License is intended to guarantee your freedom to +share and change all versions of a program--to make sure it remains free +software for all its users. We, the Free Software Foundation, use the +GNU General Public License for most of our software; it applies also to +any other work released this way by its authors. You can apply it to +your programs, too. + + When we speak of free software, we are referring to freedom, not +price. Our General Public Licenses are designed to make sure that you +have the freedom to distribute copies of free software (and charge for +them if you wish), that you receive source code or can get it if you +want it, that you can change the software or use pieces of it in new +free programs, and that you know you can do these things. + + To protect your rights, we need to prevent others from denying you +these rights or asking you to surrender the rights. Therefore, you have +certain responsibilities if you distribute copies of the software, or if +you modify it: responsibilities to respect the freedom of others. + + For example, if you distribute copies of such a program, whether +gratis or for a fee, you must pass on to the recipients the same +freedoms that you received. You must make sure that they, too, receive +or can get the source code. And you must show them these terms so they +know their rights. + + Developers that use the GNU GPL protect your rights with two steps: +(1) assert copyright on the software, and (2) offer you this License +giving you legal permission to copy, distribute and/or modify it. + + For the developers' and authors' protection, the GPL clearly explains +that there is no warranty for this free software. For both users' and +authors' sake, the GPL requires that modified versions be marked as +changed, so that their problems will not be attributed erroneously to +authors of previous versions. + + Some devices are designed to deny users access to install or run +modified versions of the software inside them, although the manufacturer +can do so. This is fundamentally incompatible with the aim of +protecting users' freedom to change the software. The systematic +pattern of such abuse occurs in the area of products for individuals to +use, which is precisely where it is most unacceptable. Therefore, we +have designed this version of the GPL to prohibit the practice for those +products. If such problems arise substantially in other domains, we +stand ready to extend this provision to those domains in future versions +of the GPL, as needed to protect the freedom of users. + + Finally, every program is threatened constantly by software patents. +States should not allow patents to restrict development and use of +software on general-purpose computers, but in those that do, we wish to +avoid the special danger that patents applied to a free program could +make it effectively proprietary. To prevent this, the GPL assures that +patents cannot be used to render the program non-free. + + The precise terms and conditions for copying, distribution and +modification follow. + + TERMS AND CONDITIONS + + 0. Definitions. + + "This License" refers to version 3 of the GNU General Public License. + + "Copyright" also means copyright-like laws that apply to other kinds of +works, such as semiconductor masks. + + "The Program" refers to any copyrightable work licensed under this +License. Each licensee is addressed as "you". "Licensees" and +"recipients" may be individuals or organizations. + + To "modify" a work means to copy from or adapt all or part of the work +in a fashion requiring copyright permission, other than the making of an +exact copy. The resulting work is called a "modified version" of the +earlier work or a work "based on" the earlier work. + + A "covered work" means either the unmodified Program or a work based +on the Program. + + To "propagate" a work means to do anything with it that, without +permission, would make you directly or secondarily liable for +infringement under applicable copyright law, except executing it on a +computer or modifying a private copy. Propagation includes copying, +distribution (with or without modification), making available to the +public, and in some countries other activities as well. + + To "convey" a work means any kind of propagation that enables other +parties to make or receive copies. Mere interaction with a user through +a computer network, with no transfer of a copy, is not conveying. + + An interactive user interface displays "Appropriate Legal Notices" +to the extent that it includes a convenient and prominently visible +feature that (1) displays an appropriate copyright notice, and (2) +tells the user that there is no warranty for the work (except to the +extent that warranties are provided), that licensees may convey the +work under this License, and how to view a copy of this License. If +the interface presents a list of user commands or options, such as a +menu, a prominent item in the list meets this criterion. + + 1. Source Code. + + The "source code" for a work means the preferred form of the work +for making modifications to it. "Object code" means any non-source +form of a work. + + A "Standard Interface" means an interface that either is an official +standard defined by a recognized standards body, or, in the case of +interfaces specified for a particular programming language, one that +is widely used among developers working in that language. + + The "System Libraries" of an executable work include anything, other +than the work as a whole, that (a) is included in the normal form of +packaging a Major Component, but which is not part of that Major +Component, and (b) serves only to enable use of the work with that +Major Component, or to implement a Standard Interface for which an +implementation is available to the public in source code form. A +"Major Component", in this context, means a major essential component +(kernel, window system, and so on) of the specific operating system +(if any) on which the executable work runs, or a compiler used to +produce the work, or an object code interpreter used to run it. + + The "Corresponding Source" for a work in object code form means all +the source code needed to generate, install, and (for an executable +work) run the object code and to modify the work, including scripts to +control those activities. However, it does not include the work's +System Libraries, or general-purpose tools or generally available free +programs which are used unmodified in performing those activities but +which are not part of the work. For example, Corresponding Source +includes interface definition files associated with source files for +the work, and the source code for shared libraries and dynamically +linked subprograms that the work is specifically designed to require, +such as by intimate data communication or control flow between those +subprograms and other parts of the work. + + The Corresponding Source need not include anything that users +can regenerate automatically from other parts of the Corresponding +Source. + + The Corresponding Source for a work in source code form is that +same work. + + 2. Basic Permissions. + + All rights granted under this License are granted for the term of +copyright on the Program, and are irrevocable provided the stated +conditions are met. This License explicitly affirms your unlimited +permission to run the unmodified Program. The output from running a +covered work is covered by this License only if the output, given its +content, constitutes a covered work. This License acknowledges your +rights of fair use or other equivalent, as provided by copyright law. + + You may make, run and propagate covered works that you do not +convey, without conditions so long as your license otherwise remains +in force. You may convey covered works to others for the sole purpose +of having them make modifications exclusively for you, or provide you +with facilities for running those works, provided that you comply with +the terms of this License in conveying all material for which you do +not control copyright. Those thus making or running the covered works +for you must do so exclusively on your behalf, under your direction +and control, on terms that prohibit them from making any copies of +your copyrighted material outside their relationship with you. + + Conveying under any other circumstances is permitted solely under +the conditions stated below. Sublicensing is not allowed; section 10 +makes it unnecessary. + + 3. Protecting Users' Legal Rights From Anti-Circumvention Law. + + No covered work shall be deemed part of an effective technological +measure under any applicable law fulfilling obligations under article +11 of the WIPO copyright treaty adopted on 20 December 1996, or +similar laws prohibiting or restricting circumvention of such +measures. + + When you convey a covered work, you waive any legal power to forbid +circumvention of technological measures to the extent such circumvention +is effected by exercising rights under this License with respect to +the covered work, and you disclaim any intention to limit operation or +modification of the work as a means of enforcing, against the work's +users, your or third parties' legal rights to forbid circumvention of +technological measures. + + 4. Conveying Verbatim Copies. + + You may convey verbatim copies of the Program's source code as you +receive it, in any medium, provided that you conspicuously and +appropriately publish on each copy an appropriate copyright notice; +keep intact all notices stating that this License and any +non-permissive terms added in accord with section 7 apply to the code; +keep intact all notices of the absence of any warranty; and give all +recipients a copy of this License along with the Program. + + You may charge any price or no price for each copy that you convey, +and you may offer support or warranty protection for a fee. + + 5. Conveying Modified Source Versions. + + You may convey a work based on the Program, or the modifications to +produce it from the Program, in the form of source code under the +terms of section 4, provided that you also meet all of these conditions: + + a) The work must carry prominent notices stating that you modified + it, and giving a relevant date. + + b) The work must carry prominent notices stating that it is + released under this License and any conditions added under section + 7. This requirement modifies the requirement in section 4 to + "keep intact all notices". + + c) You must license the entire work, as a whole, under this + License to anyone who comes into possession of a copy. This + License will therefore apply, along with any applicable section 7 + additional terms, to the whole of the work, and all its parts, + regardless of how they are packaged. This License gives no + permission to license the work in any other way, but it does not + invalidate such permission if you have separately received it. + + d) If the work has interactive user interfaces, each must display + Appropriate Legal Notices; however, if the Program has interactive + interfaces that do not display Appropriate Legal Notices, your + work need not make them do so. + + A compilation of a covered work with other separate and independent +works, which are not by their nature extensions of the covered work, +and which are not combined with it such as to form a larger program, +in or on a volume of a storage or distribution medium, is called an +"aggregate" if the compilation and its resulting copyright are not +used to limit the access or legal rights of the compilation's users +beyond what the individual works permit. Inclusion of a covered work +in an aggregate does not cause this License to apply to the other +parts of the aggregate. + + 6. Conveying Non-Source Forms. + + You may convey a covered work in object code form under the terms +of sections 4 and 5, provided that you also convey the +machine-readable Corresponding Source under the terms of this License, +in one of these ways: + + a) Convey the object code in, or embodied in, a physical product + (including a physical distribution medium), accompanied by the + Corresponding Source fixed on a durable physical medium + customarily used for software interchange. + + b) Convey the object code in, or embodied in, a physical product + (including a physical distribution medium), accompanied by a + written offer, valid for at least three years and valid for as + long as you offer spare parts or customer support for that product + model, to give anyone who possesses the object code either (1) a + copy of the Corresponding Source for all the software in the + product that is covered by this License, on a durable physical + medium customarily used for software interchange, for a price no + more than your reasonable cost of physically performing this + conveying of source, or (2) access to copy the + Corresponding Source from a network server at no charge. + + c) Convey individual copies of the object code with a copy of the + written offer to provide the Corresponding Source. This + alternative is allowed only occasionally and noncommercially, and + only if you received the object code with such an offer, in accord + with subsection 6b. + + d) Convey the object code by offering access from a designated + place (gratis or for a charge), and offer equivalent access to the + Corresponding Source in the same way through the same place at no + further charge. You need not require recipients to copy the + Corresponding Source along with the object code. If the place to + copy the object code is a network server, the Corresponding Source + may be on a different server (operated by you or a third party) + that supports equivalent copying facilities, provided you maintain + clear directions next to the object code saying where to find the + Corresponding Source. Regardless of what server hosts the + Corresponding Source, you remain obligated to ensure that it is + available for as long as needed to satisfy these requirements. + + e) Convey the object code using peer-to-peer transmission, provided + you inform other peers where the object code and Corresponding + Source of the work are being offered to the general public at no + charge under subsection 6d. + + A separable portion of the object code, whose source code is excluded +from the Corresponding Source as a System Library, need not be +included in conveying the object code work. + + A "User Product" is either (1) a "consumer product", which means any +tangible personal property which is normally used for personal, family, +or household purposes, or (2) anything designed or sold for incorporation +into a dwelling. In determining whether a product is a consumer product, +doubtful cases shall be resolved in favor of coverage. For a particular +product received by a particular user, "normally used" refers to a +typical or common use of that class of product, regardless of the status +of the particular user or of the way in which the particular user +actually uses, or expects or is expected to use, the product. A product +is a consumer product regardless of whether the product has substantial +commercial, industrial or non-consumer uses, unless such uses represent +the only significant mode of use of the product. + + "Installation Information" for a User Product means any methods, +procedures, authorization keys, or other information required to install +and execute modified versions of a covered work in that User Product from +a modified version of its Corresponding Source. The information must +suffice to ensure that the continued functioning of the modified object +code is in no case prevented or interfered with solely because +modification has been made. + + If you convey an object code work under this section in, or with, or +specifically for use in, a User Product, and the conveying occurs as +part of a transaction in which the right of possession and use of the +User Product is transferred to the recipient in perpetuity or for a +fixed term (regardless of how the transaction is characterized), the +Corresponding Source conveyed under this section must be accompanied +by the Installation Information. But this requirement does not apply +if neither you nor any third party retains the ability to install +modified object code on the User Product (for example, the work has +been installed in ROM). + + The requirement to provide Installation Information does not include a +requirement to continue to provide support service, warranty, or updates +for a work that has been modified or installed by the recipient, or for +the User Product in which it has been modified or installed. Access to a +network may be denied when the modification itself materially and +adversely affects the operation of the network or violates the rules and +protocols for communication across the network. + + Corresponding Source conveyed, and Installation Information provided, +in accord with this section must be in a format that is publicly +documented (and with an implementation available to the public in +source code form), and must require no special password or key for +unpacking, reading or copying. + + 7. Additional Terms. + + "Additional permissions" are terms that supplement the terms of this +License by making exceptions from one or more of its conditions. +Additional permissions that are applicable to the entire Program shall +be treated as though they were included in this License, to the extent +that they are valid under applicable law. If additional permissions +apply only to part of the Program, that part may be used separately +under those permissions, but the entire Program remains governed by +this License without regard to the additional permissions. + + When you convey a copy of a covered work, you may at your option +remove any additional permissions from that copy, or from any part of +it. (Additional permissions may be written to require their own +removal in certain cases when you modify the work.) You may place +additional permissions on material, added by you to a covered work, +for which you have or can give appropriate copyright permission. + + Notwithstanding any other provision of this License, for material you +add to a covered work, you may (if authorized by the copyright holders of +that material) supplement the terms of this License with terms: + + a) Disclaiming warranty or limiting liability differently from the + terms of sections 15 and 16 of this License; or + + b) Requiring preservation of specified reasonable legal notices or + author attributions in that material or in the Appropriate Legal + Notices displayed by works containing it; or + + c) Prohibiting misrepresentation of the origin of that material, or + requiring that modified versions of such material be marked in + reasonable ways as different from the original version; or + + d) Limiting the use for publicity purposes of names of licensors or + authors of the material; or + + e) Declining to grant rights under trademark law for use of some + trade names, trademarks, or service marks; or + + f) Requiring indemnification of licensors and authors of that + material by anyone who conveys the material (or modified versions of + it) with contractual assumptions of liability to the recipient, for + any liability that these contractual assumptions directly impose on + those licensors and authors. + + All other non-permissive additional terms are considered "further +restrictions" within the meaning of section 10. If the Program as you +received it, or any part of it, contains a notice stating that it is +governed by this License along with a term that is a further +restriction, you may remove that term. If a license document contains +a further restriction but permits relicensing or conveying under this +License, you may add to a covered work material governed by the terms +of that license document, provided that the further restriction does +not survive such relicensing or conveying. + + If you add terms to a covered work in accord with this section, you +must place, in the relevant source files, a statement of the +additional terms that apply to those files, or a notice indicating +where to find the applicable terms. + + Additional terms, permissive or non-permissive, may be stated in the +form of a separately written license, or stated as exceptions; +the above requirements apply either way. + + 8. Termination. + + You may not propagate or modify a covered work except as expressly +provided under this License. Any attempt otherwise to propagate or +modify it is void, and will automatically terminate your rights under +this License (including any patent licenses granted under the third +paragraph of section 11). + + However, if you cease all violation of this License, then your +license from a particular copyright holder is reinstated (a) +provisionally, unless and until the copyright holder explicitly and +finally terminates your license, and (b) permanently, if the copyright +holder fails to notify you of the violation by some reasonable means +prior to 60 days after the cessation. + + Moreover, your license from a particular copyright holder is +reinstated permanently if the copyright holder notifies you of the +violation by some reasonable means, this is the first time you have +received notice of violation of this License (for any work) from that +copyright holder, and you cure the violation prior to 30 days after +your receipt of the notice. + + Termination of your rights under this section does not terminate the +licenses of parties who have received copies or rights from you under +this License. If your rights have been terminated and not permanently +reinstated, you do not qualify to receive new licenses for the same +material under section 10. + + 9. Acceptance Not Required for Having Copies. + + You are not required to accept this License in order to receive or +run a copy of the Program. Ancillary propagation of a covered work +occurring solely as a consequence of using peer-to-peer transmission +to receive a copy likewise does not require acceptance. However, +nothing other than this License grants you permission to propagate or +modify any covered work. These actions infringe copyright if you do +not accept this License. Therefore, by modifying or propagating a +covered work, you indicate your acceptance of this License to do so. + + 10. Automatic Licensing of Downstream Recipients. + + Each time you convey a covered work, the recipient automatically +receives a license from the original licensors, to run, modify and +propagate that work, subject to this License. You are not responsible +for enforcing compliance by third parties with this License. + + An "entity transaction" is a transaction transferring control of an +organization, or substantially all assets of one, or subdividing an +organization, or merging organizations. If propagation of a covered +work results from an entity transaction, each party to that +transaction who receives a copy of the work also receives whatever +licenses to the work the party's predecessor in interest had or could +give under the previous paragraph, plus a right to possession of the +Corresponding Source of the work from the predecessor in interest, if +the predecessor has it or can get it with reasonable efforts. + + You may not impose any further restrictions on the exercise of the +rights granted or affirmed under this License. For example, you may +not impose a license fee, royalty, or other charge for exercise of +rights granted under this License, and you may not initiate litigation +(including a cross-claim or counterclaim in a lawsuit) alleging that +any patent claim is infringed by making, using, selling, offering for +sale, or importing the Program or any portion of it. + + 11. Patents. + + A "contributor" is a copyright holder who authorizes use under this +License of the Program or a work on which the Program is based. The +work thus licensed is called the contributor's "contributor version". + + A contributor's "essential patent claims" are all patent claims +owned or controlled by the contributor, whether already acquired or +hereafter acquired, that would be infringed by some manner, permitted +by this License, of making, using, or selling its contributor version, +but do not include claims that would be infringed only as a +consequence of further modification of the contributor version. For +purposes of this definition, "control" includes the right to grant +patent sublicenses in a manner consistent with the requirements of +this License. + + Each contributor grants you a non-exclusive, worldwide, royalty-free +patent license under the contributor's essential patent claims, to +make, use, sell, offer for sale, import and otherwise run, modify and +propagate the contents of its contributor version. + + In the following three paragraphs, a "patent license" is any express +agreement or commitment, however denominated, not to enforce a patent +(such as an express permission to practice a patent or covenant not to +sue for patent infringement). To "grant" such a patent license to a +party means to make such an agreement or commitment not to enforce a +patent against the party. + + If you convey a covered work, knowingly relying on a patent license, +and the Corresponding Source of the work is not available for anyone +to copy, free of charge and under the terms of this License, through a +publicly available network server or other readily accessible means, +then you must either (1) cause the Corresponding Source to be so +available, or (2) arrange to deprive yourself of the benefit of the +patent license for this particular work, or (3) arrange, in a manner +consistent with the requirements of this License, to extend the patent +license to downstream recipients. "Knowingly relying" means you have +actual knowledge that, but for the patent license, your conveying the +covered work in a country, or your recipient's use of the covered work +in a country, would infringe one or more identifiable patents in that +country that you have reason to believe are valid. + + If, pursuant to or in connection with a single transaction or +arrangement, you convey, or propagate by procuring conveyance of, a +covered work, and grant a patent license to some of the parties +receiving the covered work authorizing them to use, propagate, modify +or convey a specific copy of the covered work, then the patent license +you grant is automatically extended to all recipients of the covered +work and works based on it. + + A patent license is "discriminatory" if it does not include within +the scope of its coverage, prohibits the exercise of, or is +conditioned on the non-exercise of one or more of the rights that are +specifically granted under this License. You may not convey a covered +work if you are a party to an arrangement with a third party that is +in the business of distributing software, under which you make payment +to the third party based on the extent of your activity of conveying +the work, and under which the third party grants, to any of the +parties who would receive the covered work from you, a discriminatory +patent license (a) in connection with copies of the covered work +conveyed by you (or copies made from those copies), or (b) primarily +for and in connection with specific products or compilations that +contain the covered work, unless you entered into that arrangement, +or that patent license was granted, prior to 28 March 2007. + + Nothing in this License shall be construed as excluding or limiting +any implied license or other defenses to infringement that may +otherwise be available to you under applicable patent law. + + 12. No Surrender of Others' Freedom. + + If conditions are imposed on you (whether by court order, agreement or +otherwise) that contradict the conditions of this License, they do not +excuse you from the conditions of this License. If you cannot convey a +covered work so as to satisfy simultaneously your obligations under this +License and any other pertinent obligations, then as a consequence you may +not convey it at all. For example, if you agree to terms that obligate you +to collect a royalty for further conveying from those to whom you convey +the Program, the only way you could satisfy both those terms and this +License would be to refrain entirely from conveying the Program. + + 13. Use with the GNU Affero General Public License. + + Notwithstanding any other provision of this License, you have +permission to link or combine any covered work with a work licensed +under version 3 of the GNU Affero General Public License into a single +combined work, and to convey the resulting work. The terms of this +License will continue to apply to the part which is the covered work, +but the special requirements of the GNU Affero General Public License, +section 13, concerning interaction through a network will apply to the +combination as such. + + 14. Revised Versions of this License. + + The Free Software Foundation may publish revised and/or new versions of +the GNU General Public License from time to time. Such new versions will +be similar in spirit to the present version, but may differ in detail to +address new problems or concerns. + + Each version is given a distinguishing version number. If the +Program specifies that a certain numbered version of the GNU General +Public License "or any later version" applies to it, you have the +option of following the terms and conditions either of that numbered +version or of any later version published by the Free Software +Foundation. If the Program does not specify a version number of the +GNU General Public License, you may choose any version ever published +by the Free Software Foundation. + + If the Program specifies that a proxy can decide which future +versions of the GNU General Public License can be used, that proxy's +public statement of acceptance of a version permanently authorizes you +to choose that version for the Program. + + Later license versions may give you additional or different +permissions. However, no additional obligations are imposed on any +author or copyright holder as a result of your choosing to follow a +later version. + + 15. Disclaimer of Warranty. + + THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY +APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT +HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY +OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, +THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR +PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM +IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF +ALL NECESSARY SERVICING, REPAIR OR CORRECTION. + + 16. Limitation of Liability. + + IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING +WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS +THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY +GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE +USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF +DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD +PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), +EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF +SUCH DAMAGES. + + 17. Interpretation of Sections 15 and 16. + + If the disclaimer of warranty and limitation of liability provided +above cannot be given local legal effect according to their terms, +reviewing courts shall apply local law that most closely approximates +an absolute waiver of all civil liability in connection with the +Program, unless a warranty or assumption of liability accompanies a +copy of the Program in return for a fee. + + END OF TERMS AND CONDITIONS + + How to Apply These Terms to Your New Programs + + If you develop a new program, and you want it to be of the greatest +possible use to the public, the best way to achieve this is to make it +free software which everyone can redistribute and change under these terms. + + To do so, attach the following notices to the program. It is safest +to attach them to the start of each source file to most effectively +state the exclusion of warranty; and each file should have at least +the "copyright" line and a pointer to where the full notice is found. + + + Copyright (C) + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + +Also add information on how to contact you by electronic and paper mail. + + If the program does terminal interaction, make it output a short +notice like this when it starts in an interactive mode: + + Copyright (C) + This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. + This is free software, and you are welcome to redistribute it + under certain conditions; type `show c' for details. + +The hypothetical commands `show w' and `show c' should show the appropriate +parts of the General Public License. Of course, your program's commands +might be different; for a GUI interface, you would use an "about box". + + You should also get your employer (if you work as a programmer) or school, +if any, to sign a "copyright disclaimer" for the program, if necessary. +For more information on this, and how to apply and follow the GNU GPL, see +. + + The GNU General Public License does not permit incorporating your program +into proprietary programs. If your program is a subroutine library, you +may consider it more useful to permit linking proprietary applications with +the library. If this is what you want to do, use the GNU Lesser General +Public License instead of this License. But first, please read +. diff --git a/PepSeqPred_logo_black.png b/PepSeqPred_logo_black.png new file mode 100644 index 0000000..cf8c797 Binary files /dev/null and b/PepSeqPred_logo_black.png differ diff --git a/PepSeqPred_logo_white.png b/PepSeqPred_logo_white.png new file mode 100644 index 0000000..472250a Binary files /dev/null and b/PepSeqPred_logo_white.png differ diff --git a/README.md b/README.md index 503bd3b..8425d8e 100644 --- a/README.md +++ b/README.md @@ -1 +1,421 @@ -# PepSeqPred \ No newline at end of file +

+ + + + PepSeqPred logo + +

+ +# Overview + +PepSeqPred is a residue-level epitope prediction pipeline for protein workflows. +It converts upstream assay and sequence data into training-ready artifacts, trains feed-forward neural network models on ESM-2 embeddings, and produces binary residue masks for downstream inference. + +## About + +PepSeqPred is designed for research workflows where you need to: +- map peptide-level signals to residue-level supervision, +- train reproducible epitope prediction models, +- run large jobs on HPCs with DistributedDataParallel (DDP), +- generate per-residue binary epitope predictions to develop new peptide libraries. + +The pipeline is built around CLI entrypoints in `src/pepseqpred/apps/` and matching HPC scripts in `scripts/hpc/`. + +## What Goes In / What Comes Out + +Typical inputs: +- Metadata and reactivity tables (TSV) for preprocessing and label generation. +- Protein FASTA files for embedding generation and prediction. +- Optional metadata file for ID-family embedding key generation. + +Typical outputs: +- Preprocessed model input table (TSV). +- Per-protein ESM-2 embedding shards (`.pt`) plus embedding index CSV. +- Residue-level label shards (`.pt`) including epitope / uncertain / non-epitope supervision. +- FFNN checkpoints and run summaries (for standard training or Optuna tuning). +- Predicted binary epitope mask FASTA files for inference. + +## Pipeline Snapshot + +```text +Preprocess data + -> cleaned + labeled metadata TSV for downstream steps + +Generate ESM-2 embeddings + -> per-protein embedding .pt files (+ index CSV), often sharded + +Generate residue-level labels + -> label shard .pt files aligned to embedding keys/shards + +Train FFNN (DDP) + -> checkpoints + metrics artifacts + +Predict epitopes + -> output FASTA with binary residue mask predictions + +Optional: Optuna hyperparameter tuning + -> study storage + trial CSV + best-model artifacts +``` + +## Project Scope + +This repository supports both: +- local development and validation of pipeline logic, and +- production-scale HPC execution for embedding generation, training, and tuning. + +## Maintainer Contact + +- Use GitHub issues for normal development questions, bug reports, and feature requests. +- Use email for private or sensitive matters that should not be posted publicly. +- Maintainer contact: [Jeffrey Hoelzel](mailto:jmh2338@nau.edu) or [Jason Ladner](mailto:jason.ladner@nau.edu). + +## Prerequisites + +Software: +- At least Python `3.12` (required by project configuration and CI). +- `pip` and virtual environments (`venv`) or `conda`. +- `git` for cloning and contribution workflows. + +Platform notes: +- Local development can run on CPU. +- GPU is highly recommended for embedding generation and required for practical training/tuning runtimes. +- HPC scripts in `scripts/hpc/` assume a SLURM-style environment and module-based setup (for example, `anaconda3` and `cuda`). + +## Setup + +### Option A: Conda Environment (recommended) + +Local CPU-oriented environment: + +```bash +conda env create -f envs/environment.local.yml +conda activate pepseqpred +pip install -e .[dev] +``` + +HPC/GPU-oriented environment: + +```bash +conda env create -f envs/environment.hpc.yml +conda activate pepseqpred +pip install -e .[dev] +``` + +### Option B: Pip + Virtual Environment + +Linux/macOS: + +```bash +python3 -m venv .venv +source .venv/bin/activate +python -m pip install --upgrade pip +pip install -e .[dev] +``` + +Windows PowerShell: + +```powershell +python -m venv .venv +.venv\Scripts\Activate.ps1 +python -m pip install --upgrade pip +pip install -e .[dev] +``` + +CI-equivalent install shortcut: + +```bash +pip install -r requirements.txt +``` + +## Verify Setup + +Confirm package and CLI entrypoints: + +```bash +python -c "import pepseqpred; print('pepseqpred import ok')" +pepseqpred-preprocess --help +pepseqpred-esm --help +pepseqpred-train-ffnn --help +pepseqpred-predict --help +``` + +Run required preflight checks before any development or pipeline usage: + +```bash +ruff check . +pytest -m "unit or integration or e2e" +``` + +This repository expects all of the checks above to pass before you start development work or run pipeline stages. + +## Pipeline Stages and Hardware Expectations + +Run the main pipeline in this order: +1. Preprocess metadata/reactivity data. +2. Generate ESM-2 embeddings. +3. Generate residue-level labels. +4. Train FFNN model. +5. Predict epitopes on new FASTA input. + +Optional branch: +1. Run Optuna tuning after label generation instead of fixed-parameter FFNN training. + +### Hardware Matrix (default `scripts/hpc/` settings) + +| Stage | Hardware target | Default SLURM request in repo scripts | +| --- | --- | --- | +| Preprocess | CPU only | Local shell helper (no fixed `#SBATCH` resources) | +| Embeddings | GPU | `a100` (1 GPU), `2` CPU/GPU, `8G`/GPU, `01:00:00`, array `0-3` | +| Labels | CPU only | `1` CPU, `16G` RAM, `01:00:00` | +| Train FFNN | Multi-GPU | `4x a100`, `20` CPU, `256G` RAM, `12:00:00` | +| Train FFNN Optuna | Multi-GPU | `4x a100`, `20` CPU, `448G` RAM, `48:00:00` | +| Predict | GPU | `a100` (1 GPU), `4` CPU, `32G` RAM, `00:30:00` | + +These are baseline defaults from the current HPC scripts, not strict requirements for every dataset size. You will need to increase or decrease SLURM requests depending on the hardware available to you. + +### Stage 1: Preprocess Metadata and Reactivity Inputs + +Purpose: +- Merge metadata with z-score reactivity data and generate training-ready labels at preprocessing stage. + +Required inputs: +- Metadata TSV (for example, PV1 metadata). +- Reactivity/z-score TSV. + +Local CLI example: + +```bash +pepseqpred-preprocess data/meta.tsv data/zscores.tsv --save +``` + +Expected outputs: +- A generated TSV in the working directory, named like `input_data____.tsv`. + +Expected hardware: +- CPU only; lightweight compared with downstream stages. + +### Stage 2: Generate ESM-2 Embeddings + +Purpose: +- Convert protein FASTA sequences into per-residue ESM-2 embeddings. + +Required inputs: +- FASTA file. +- Metadata file when using `id-family` embedding keys (default mode). + +Local CLI example: + +```bash +pepseqpred-esm \ + --fasta-file data/targets.fasta \ + --metadata-file data/targets.metadata \ + --embedding-key-mode id-family \ + --key-delimiter - \ + --model-name esm2_t33_650M_UR50D \ + --max-tokens 1022 \ + --batch-size 8 \ + --out-dir localdata/esm2_run +``` + +HPC script example: + +```bash +sbatch --export=ALL,IN_FASTA=/scratch/$USER/data/targets.fasta scripts/hpc/generateembeddings.sh +``` + +Expected outputs: +- Per-sequence `.pt` embeddings under `/artifacts/pts/` (or shard subfolders in sharded mode). +- Embedding index CSV under `/artifacts/`. + +Expected hardware: +- GPU strongly recommended; this is the first expensive stage. + +### Stage 3: Generate Residue-Level Label Shards + +Purpose: +- Build dense residue labels aligned to generated embedding keys/shards. + +Required inputs: +- Preprocessed metadata TSV. +- Embedding directory (or shard directories). + +Local CLI example: + +```bash +pepseqpred-labels \ + data/input_data_20_4_10_all.tsv \ + localdata/labels/labels_shard_000.pt \ + --emb-dir localdata/esm2_run/artifacts/pts/shard_000 \ + --restrict-to-embeddings \ + --calc-pos-weight \ + --embedding-key-delim - +``` + +HPC script examples: + +```bash +sbatch --array=0-3 scripts/hpc/generatelabels.sh data/input_data_20_4_10_all.tsv /scratch/$USER/labels /scratch/$USER/esm2/artifacts/pts +``` + +Expected outputs: +- Label shard files such as `labels_shard_000.pt`. +- Optional positive class weight in the saved payload when `--calc-pos-weight` is used. + +Expected hardware: +- CPU only; moderate memory. + +### Stage 4: Train FFNN + +Purpose: +- Train PepSeqPred FFNN on embedding shards and label shards. + +Required inputs: +- One or more embedding shard directories. +- One or more label shard `.pt` files aligned to those embeddings. + +Local smoke-test CLI example (small subset): + +```bash +pepseqpred-train-ffnn \ + --embedding-dirs localdata/esm2_run/artifacts/pts/shard_000 \ + --label-shards localdata/labels/labels_shard_000.pt \ + --epochs 1 \ + --subset 100 \ + --save-path localdata/models/ffnn_smoke \ + --results-csv localdata/models/ffnn_smoke/runs.csv +``` + +HPC script example: + +```bash +sbatch scripts/hpc/trainffnn.sh \ + /scratch/$USER/esm2/artifacts/pts/shard_000 /scratch/$USER/esm2/artifacts/pts/shard_001 /scratch/$USER/esm2/artifacts/pts/shard_002 /scratch/$USER/esm2/artifacts/pts/shard_003 \ + -- \ + /scratch/$USER/labels/labels_shard_000.pt /scratch/$USER/labels/labels_shard_001.pt /scratch/$USER/labels/labels_shard_002.pt /scratch/$USER/labels/labels_shard_003.pt +``` + +Expected outputs: +- Run directories under `--save-path` containing checkpoints (for example `fully_connected.pt`). +- Multi-run CSV summary (default `multi_run_results.csv`, or `--results-csv` path). +- Aggregated `multi_run_summary.json`. + +Expected hardware: +- Practical training is multi-GPU/HPC-oriented. + +### Stage 5 (Optional): Train FFNN with Optuna + +Purpose: +- Run distributed hyperparameter optimization over FFNN architecture/training ranges. + +Required inputs: +- Same embedding shard directories and label shard files used by training. + +HPC script example: + +```bash +sbatch scripts/hpc/trainffnnoptuna.sh \ + /scratch/$USER/esm2/artifacts/pts/shard_000 /scratch/$USER/esm2/artifacts/pts/shard_001 /scratch/$USER/esm2/artifacts/pts/shard_002 /scratch/$USER/esm2/artifacts/pts/shard_003 \ + -- \ + /scratch/$USER/labels/labels_shard_000.pt /scratch/$USER/labels/labels_shard_001.pt /scratch/$USER/labels/labels_shard_002.pt /scratch/$USER/labels/labels_shard_003.pt +``` + +Expected outputs: +- Trial metrics CSV (`--csv-path`). +- Optuna storage DB (`--storage`), SQLite on scratch. +- Trial checkpoint directories under `--save-path`. +- Best-trial metadata JSON under `--save-path`. + +Expected hardware: +- Most expensive stage; budget multi-GPU runtime from several hours to days depending on `--n-trials`. + +### Stage 6: Predict Residue-Level Epitopes + +Purpose: +- Apply trained checkpoint to new FASTA input and emit binary residue masks. + +Required inputs: +- Trained checkpoint `.pt`. +- Input FASTA file. + +Local CLI example: + +```bash +pepseqpred-predict \ + localdata/models/ffnn_smoke/run_001_split_11_train_101/fully_connected.pt \ + data/inference_targets.fasta \ + --output-fasta localdata/predictions/predictions.fasta \ + --model-name esm2_t33_650M_UR50D \ + --max-tokens 1022 +``` + +HPC script example: + +```bash +sbatch scripts/hpc/predictepitope.sh /scratch/$USER/models/ffnn_v1/run_001_split_11_train_101/fully_connected.pt /scratch/$USER/data/inference_targets.fasta /scratch/$USER/predictions/predictions.fasta +``` + +Expected outputs: +- Output FASTA with binary residue-level mask predictions. +- Prediction logs (console and optional log directory). + +Expected hardware: +- Single GPU is recommended for throughput; CPU inference is possible but much slower. + +### Stage Compatibility Notes + +- Keep embedding key scheme consistent (`id` vs `id-family`) between embedding and label generation. +- Keep shard alignment explicit: embedding shard directories should map cleanly to label shard files. +- Use local smoke settings (`--subset`, low epochs, single shard) before submitting expensive HPC jobs. + +## Reproducibility and Output Conventions + +- Use a new output root per run/study to avoid accidental overwrite of prior artifacts. +- Keep preprocessing outputs, embeddings, labels, checkpoints, and predictions in separate subdirectories. +- Keep `split-seeds`, `train-seeds`, and core hyperparameters fixed when comparing experiments. +- Keep `split-type` and embedding key scheme (`id` vs `id-family`) unchanged within a single experiment. +- Do not mix artifacts from different preprocessing thresholds into one training run. + +Suggested layout: + +```text +localdata/ + runs/ + / + preprocess/ + embeddings/ + labels/ + models/ + predictions/ + logs/ +``` + +## Troubleshooting + +Common issues and fixes: + +- `Metadata file is required for --embedding-key-mode='id-family'`: + Provide `--metadata-file`, or use `--embedding-key-mode id` if that is your intended key scheme. +- `--embedding-key-delim must be '' or '-'`: + Use `-` for `ID-family.pt` naming and empty delimiter for `ID.pt` naming. +- `No .pt files found in ` during label generation: + Verify Stage 2 completed successfully and `--emb-dir` points to the directory that directly contains embedding `.pt` files. +- Label shard missing `class_stats` when training: + Rebuild labels with `--calc-pos-weight`, or pass `--pos-weight` explicitly to training. +- `--hidden-sizes and --dropouts must be the same length`: + Ensure both CSV lists have one value per hidden layer. +- Prediction threshold errors (`(0.0, 1.0)` required): + Set `--threshold` strictly between `0` and `1`, or omit it to use checkpoint/default behavior. +- DDP or multi-GPU runs stall/hang: + Confirm the requested GPU count matches `torchrun --nproc_per_node`, and validate with a small single-shard smoke test first. +- CUDA OOM: + Reduce embedding `--batch-size`, reduce training `--batch-size`, or lower trial/search scope for Optuna. + +## Pre-Run Checklist + +Before starting a real run: + +- Environment is activated and `pip install -e .[dev]` completed. +- Required preflight checks pass: `ruff check .` and `pytest -m "unit or integration or e2e"`. +- Stage input files exist and are from the intended experiment branch. +- Output paths are unique for this run and will not overwrite prior artifacts. +- Embedding key scheme/delimiter and split configuration are consistent across stages. diff --git a/scripts/hpc/generatelabels.sh b/scripts/hpc/generatelabels.sh index db8db1b..daa28a8 100644 --- a/scripts/hpc/generatelabels.sh +++ b/scripts/hpc/generatelabels.sh @@ -4,7 +4,7 @@ #SBATCH --ntasks=1 #SBATCH --cpus-per-task=1 #SBATCH --mem=16G -#SBATCH --time=1:30:00 +#SBATCH --time=1:00:00 #SBATCH --output=/scratch/%u/labels_slurm/%A_%a/%x.out #SBATCH --error=/scratch/%u/labels_slurm/%A_%a/%x.err