Skip to content

build: releasing first beta version v0.1.0b0 - #185

Merged
olive005 merged 11 commits into
mainfrom
develop
Jul 10, 2026
Merged

build: releasing first beta version v0.1.0b0#185
olive005 merged 11 commits into
mainfrom
develop

Conversation

@olive005

@olive005 olive005 commented Jul 10, 2026 •
edited by chrbrunk
Loading

Copy link
Copy Markdown
Collaborator

This PR contains several changes related to the first beta release of alf.

A few changes like the introduction of Colab notebook links needs to be tested when the repo is public, before merging.

chrbrunk and others added 4 commits July 10, 2026 10:45
@chrbrunk @surana01
ci: Add release pipeline to CI (#169)
30f1d4d 
* ci: add ci release job and introduce small bug in top-level pyproject.toml to verify the version check

* ci: fix early-failing release ci job

* ci: fix linters

* chore: add init files to packages (#179)

* Update release.yml

* change version to a0

* Update release.yml

* revert version

* build: bump version to beta version b0

---------

Co-authored-by: Shikha Surana <114497416+surana01@users.noreply.github.com>
Co-authored-by: Shikha <s.surana@instadeep.com>
@olive005
fix: linting
369818d 
Co-Authored-By: olive005 <o.gallup@instadeep.com>
@olive005 @claude
Final release prep: Colab badges + public CI runners (#181)
505fc9c 
* docs: improve tutorial notebooks (setup, offline/online fixes, explanations)

Address reviewer feedback across the tutorial notebooks:

- Add a standard, aligned "Setup" section + optional commented `%pip install`
  box to every tutorial so they work in both dev mode and a plain pip install.
- tutorials/pyproject.toml: drop the now-unused pyrosetta-installer dep and add
  esm2/chemprop/guacamol/protein/molecule dependency-groups mirroring the root,
  so `uv sync --group <name>` works from tutorials/ (fixes the ESM-2 group error).
- Offline design: Experiment Overview numbering -> bullets; note the surrogate
  is not pre-trained; add Greedy-vs-advanced-acquisition note; reduce logging;
  fix Step 10 metrics (iloc[-2], iloc[1:-1], use the DataFrame index, no "round"
  column); add a results-interpretation paragraph.
- Online design: align Step 0 setup with offline; replace the ~30-min PyRosetta
  oracle with a fast, pip-installable in-silico oracle (a CNNModel trained on the
  full GFP dataset via Oracle(scorer=...)); mutate over the DNA alphabet to match
  the GFP nucleotide sequences; apply the same Step 10 metric/plot fixes; add
  results interpretation.
- GP/CNN/Ensemble: more inter-cell explanation and results commentary; ensemble
  and guacamol silence RDKit fingerprint-API deprecation noise at import.
- ESM-2: link the Installation Guide, GPU note, CPU-time note on PLL scoring, and
  a note that the small/short-training validation plot is illustrative.
- Chemprop: document the required chemprop extra.
- ProteinGym: add ProteinGym link + citation, recommend the offline tutorial
  first, and drop the NaN round-0/summary rows when printing/plotting.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* chore: apply pre-commit (ruff E402 noqa, nbstripout, eof) to tutorials

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix: address PR review on tutorials (online flat-curve note, round count, repro)

- online: explain the (expected) flat fitness curve honestly — single-nucleotide
  moves give the CNN oracle near-constant signal; reframe as a mechanics demo and
  point to the offline tutorial for the optimisation payoff.
- online: seed torch before training the in-silico oracle for reproducibility;
  read metrics via save_path (consistent with offline).
- offline + online: fix off-by-one in the printed "Total Rounds" count (the
  trailing experiment-summary row was being counted as a round).

Both notebooks re-executed end-to-end and pass; pre-commit clean.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* docs: add explanatory paragraphs above code blocks and below plots

Address the reviewer's storytelling rule of thumb across the lighter tutorials:
a short paragraph above every code block explaining what/why, and a paragraph
below every plot interpreting the result. Filled gaps in guacamol, ensemble,
esm2, gp, proteingym, chemprop, cnn and the two design tutorials. Markdown-only
additions — code cells are unchanged from the previous commit, so all notebooks
still execute as before.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* docs: interpret remaining plots and introduce proteingym component steps

Close the residual storytelling gaps a reviewer flagged: GP's first 1D fit and
final 2D field plots now have explicit uncertainty interpretation (the core GP
lesson), the GuacaMol correlation heatmap and aspirin-overlay figure are
interpreted, and ProteinGym Steps 5 and 7 (acquisition, optimizer) gain
explanatory prose. Markdown-only; code cells unchanged, all notebooks still run.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* docs: British spelling in tutorial prose and comments (team-review)

Apply the repo's British-spelling convention across the tutorial markdown,
comments and print messages: analyse, visualise, initialise, optimisation,
behaviour. Limited to prose/comments/print-strings — code identifiers and
matplotlib kwargs (Optimizer, optimizer, color=) are untouched, so execution
is unchanged.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* chore: move seaborn from default tutorial deps to the guacamol group

seaborn is only used by the GuacaMol tutorial, so it doesn't belong in the
default `[project.dependencies]`. Move it into the `guacamol` group alongside
`alf_tools[guacamol]`; `uv sync --group guacamol` (and the notebook's optional
pip box) still provide it. Verified the GuacaMol tutorial runs end-to-end after
the change.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* Revert "chore: move seaborn from default tutorial deps to the guacamol group"

Keep seaborn as a default tutorial dependency, per preference.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* docs: MLIP-style onboarding — live pip-install cell + Open-in-Colab badges

Make every tutorial open-and-run for a new user with no repo clone (inspired by
the MLIP tutorials):

- Add an "Open in Colab" badge under each title.
- Replace the dev-mode-first Setup with a Setup that LEADS with a live (uncommented)
  `%pip install` cell installing ALF (+ that tutorial's deps) from the git repo;
  dev-mode (clone + uv sync) is now a one-line "skip this cell" note.
- CPU PyTorch by default with a commented GPU line; a note that this collapses to
  `pip install alf_tools`/`alf_core` once ALF is on PyPI. Tracks `main` for now
  (version tags coming soon).

Per-notebook install targets derived from actual imports (core vs tools, esm2/
chemprop/guacamol extras, matplotlib/pandas/seaborn/umap). Validated the new path
end-to-end in a fresh venv (quickstart installed alf_core from git and ran);
notebook bodies are unchanged from the previous commit and all 16 still pass the
paragraph audit. pre-commit clean.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix: install alf_core from git in tools-based tutorials (PyPI stub is a placeholder)

`alf_tools` depends on `alf_core`, and the only `alf-core` on PyPI is an empty
0.0.0 placeholder, so `%pip install "...#subdirectory=tools"` alone resolves the
stub and the notebook then fails at `import alf_core`. Prepend an explicit
`"alf_core @ git+...#subdirectory=core"` to every tools-based install cell so pip
uses the real package from git (mirrors the two-step install in the docs).

Verified in a fresh venv: installing both from git imports alf_core, Candidate,
CNNModel and GFP successfully. Collapses to `pip install alf_tools[...]` once ALF
is published to PyPI. Core-only notebooks are unaffected.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* docs: make tutorial result analysis robust to the metrics.csv/summary.csv split (#159)

Rework Step 10 in the offline, online and proteingym tutorials to select acquisition
rounds via metrics.dropna(subset=["acquired_candidates/round_mean"]) instead of positional
iloc. This drops the initial round-0 row and any trailing per-experiment summary row, so the
analysis is correct whether summary metrics live in metrics.csv (current) or a separate
summary.csv (PR #159). X-axis uses the `round` column when present, else the index.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* docs: graft #151 code-cell splitting into model notebooks; MNIST quickstart onboarding

Post-merge reconciliation of PR #151:
- gp/cnn/ensemble/chemprop: kept our versions (rich prose, plot interpretations, Colab
  onboarding, British spelling) and grafted #151's finer code-cell splitting
  (semantics-preserving: executable code byte-identical, only cell boundaries + intros added).
- alf_core_quickstart (now #151's MNIST demo): re-applied our onboarding — Open-in-Colab
  badge + live `%pip install` (alf-core git URL) replacing the static install note.
- nbstripout/eof normalisation on the #151 extending_base_classes notebooks.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* docs: guard empty acquisition metrics and note summary.csv (review fixes)

Address LOW review findings on the Step 10 analysis cells:
- offline + online: raise a clear ValueError if metrics.csv has no acquisition
  rounds, before indexing the acquired-batch metrics with .iloc[-1].
- offline: note that per-round metrics are in metrics.csv while per-experiment
  aggregates (e.g. auc_top_k) are written to summary.csv (per #159).

Verified: offline and online run end-to-end against #159's logger, no NaN.

Co-Authored-By: olive005 <o.gallup@instadeep.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix: move proteingym

* docs: note Colab availability under the badge; fix proteingym links after move

- Add a caption under every Open-in-Colab badge: it only works once ALF is
  public / on PyPI, so internal users aren't surprised by a 404 today.
- proteingym moved experiments/ -> datasets/: repoint its own badge URL to the
  datasets/ path and fix its links to the offline tutorial (now ../experiments/).

Co-Authored-By: olive005 <o.gallup@instadeep.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* docs: add PyPI-swap TODO above each git install + Colab runtime note

- Add a per-tutorial `# TODO(pypi)` comment above each git-based `%pip install`
  giving the exact PyPI command to switch to once ALF is published, e.g.
  `%pip install "alf_tools[esm2]" matplotlib pandas umap-learn` (derived from
  each notebook's actual deps; core-only tutorials use `alf_core`).
- Add a short Colab note to each Setup section (first install may take a few
  minutes; restart the runtime if prompted; pick a GPU runtime + uncomment the
  GPU line). Core-only notebooks omit the GPU hint.

torch is left as a default dependency (alf_tools requires it; pulled in
automatically). Heavy extras keep the standard extras-group install — no
special optional-install handling added.

Co-Authored-By: olive005 <o.gallup@instadeep.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* docs: list the ProteinGym tutorial in the tutorials index

Add the ProteinGym tutorial under "Go deeper on datasets" (it lives in
tutorials/datasets/ after the move) — a discoverability gap that pre-dated the
move. Links to the notebook and notes it mirrors the offline design loop.

Co-Authored-By: olive005 <o.gallup@instadeep.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* chore: remove Colab badges, captions and Colab notes from tutorials

The Colab-specific onboarding (Open-in-Colab badge, its caption, the Colab
runtime note, and "or on Google Colab" phrasing) is being split into a separate
PR (branch feat/colab-badge -> improve/tutorials). The live pip-install
onboarding, PyPI TODOs and links stay here.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Co-Authored-By: olive005 <o.gallup@instadeep.com>

* feat: add Open-in-Colab badges and Colab guidance to tutorials

Re-adds the Colab-specific onboarding split out of improve/tutorials: the
"Open in Colab" badge + caption on each notebook, the Colab runtime note
(restart / GPU hint), and the "or on Google Colab" setup phrasing. This branch
targets improve/tutorials so the Colab support lands as its own PR.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Co-Authored-By: olive005 <o.gallup@instadeep.com>

* fix: naturalise writing, american --> britix spelling

Co-Authored-By: olive005 <o.gallup@instadeep.com>

* docs: drop TODO(pypi) comments from tutorial install cells

The eventual git-install -> PyPI swap is tracked in #175, so the per-notebook
TODO(pypi) scaffolding is redundant. The live git-based %pip install and the
optional GPU line are unchanged.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Co-Authored-By: olive005 <o.gallup@instadeep.com>

* ci: switch tests_and_linters runners to ubuntu-latest for public repo

Migrates the linters and tests jobs off the private instadeep-ci /
instadeep-ci-4 runners now that the repo is going public, per issue #175.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CPprWaGCsxTbZFPyG6JD3s

* chore: use PyPI installs in tutorial notebooks

Replace the git+https ALF installs in tutorial notebooks with their PyPI
package names (alf_core / alf_tools), keeping the tools extras (e.g.
alf_tools[esm2]). PyPI is not published yet, but the notebooks now point
at the intended install path.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: olive005 <o.gallup@instadeep.com>

* fix: separate args

Co-Authored-By: olive005 <o.gallup@instadeep.com>

* fix: update nbstripout

Co-Authored-By: olive005 <o.gallup@instadeep.com>

* ci: show diff on pre-commit failure

Add --show-diff-on-failure to the pre-commit run so hooks that rewrite
files in place (e.g. nbstripout) print the exact diff of what changed,
making CI lint failures diagnosable instead of a bare "files were
modified by this hook".

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: olive005 <o.gallup@instadeep.com>

* fix: change kernelspec; This prevents nbstripout from preserving language_info.version and other volatile fields which cause diffs across environments.

Co-Authored-By: olive005 <o.gallup@instadeep.com>

* fix: linting in pre commit yaml

Co-Authored-By: olive005 <o.gallup@instadeep.com>

* fix: rm nbstripout hook

Co-Authored-By: olive005 <o.gallup@instadeep.com>

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
@olive005
Merge branch 'main' into develop
c5ba0e1
@chrbrunk chrbrunk changed the title Develop build: releasing first beta version v0.1.0b0 Jul 10, 2026
@github-actions

github-actions Bot commented Jul 10, 2026 •
edited
Loading

Copy link
Copy Markdown

Coverage

Core Coverage Report
FileStmtsMissCoverMissing
core/alf_core
   __init__.py170100% 
core/alf_core/dataclasses
   __init__.py90100% 
   candidate.py70987%37–38, 87–88, 113, 115, 134, 143, 145
   labelled_candidates.py61296%213–214
   predictions.py36197%74
   results.py28196%83
   round_metrics.py90100% 
   state.py250100% 
   surrogate_epoch_metrics.py130100% 
core/alf_core/dataset
   base_dataset.py128794%112, 211, 328, 338, 340, 345, 353
   splitting_utils.py72888%64–65, 70, 202, 207, 214, 236, 238
core/alf_core/model
   base_model.py38586%76, 91, 104, 116, 154
   normaliser.py820100% 
core/alf_core/optimizer
   acquisition_function.py6183%39
   optimizer.py320100% 
   search.py451175%43, 55, 71, 83, 95, 117, 160, 172, 185–186, 199
core/alf_core/oracle
   oracle.py21385%72–74
core/alf_core/surrogate
   surrogate.py200100% 
core/alf_core/tasks
   base_task.py28292%66, 87
   design_task.py42197%116
   supervised_task.py210100% 
   zeroshot_task.py150100% 
core/alf_core/utils
   enums.py50100% 
   state_logger.py85989%64, 78, 119–120, 192, 226–227, 292, 311
core/alf_core/utils/metrics
   __init__.py60100% 
   acquisition_batch.py59493%38, 119–120, 131
   aggregate.py41197%158
   base.py510100% 
   calibration.py220100% 
   classification.py54688%176–177, 180, 183–185
   regression.py1490100% 
TOTAL12907194% 

Tests Skipped Failures Errors Time
396 0 💤 0 ❌ 0 🔥 3.837s ⏱️

@github-actions

github-actions Bot commented Jul 10, 2026 •
edited
Loading

Copy link
Copy Markdown

Coverage

Tools Coverage Report
FileStmtsMissCoverMissing
tools/alf_tools
   __init__.py20100% 
tools/alf_tools/datasets
   __init__.py7271%27–28
   flip.py96297%199, 307
   gfp.py28196%61
   proteingym.py118595%71, 198, 226, 283, 294
tools/alf_tools/datasets/guacamol
   __init__.py30100% 
   guacamol_dataset.py1692088%118, 181–182, 185–187, 284–289, 295–299, 354, 372, 383
   guacamol_scoring.py209199%209
   guacamol_utils.py1221091%216–220, 230, 240–241, 370–371
tools/alf_tools/models
   __init__.py40880%54–55, 65–66, 78–79, 91–92
   chemprop.py185597%185, 278, 322, 403–404
   cnn.py227796%290, 386, 436, 518, 522, 545, 656
   ensemble.py1301588%98, 112, 116, 189, 246–247, 253, 255–256, 276, 283, 344–345, 347–348
   esm2.py3432094%34–35, 76, 93, 164, 184, 195, 211, 307, 352, 362, 518, 526, 539, 597, 655, 663, 669, 707, 843
   esmfold.py112595%25–26, 123–124, 255
   gp.py2762092%128, 324, 362, 405, 412–413, 415, 465, 470, 542–544, 546, 548–550, 589, 660, 831, 928
   mlip.py1382681%101, 115, 123, 136, 209, 232–234, 269, 271, 298–300, 302–304, 311–312, 343, 363–364, 366–368, 370–371
   mlp.py204498%272, 307, 460, 510
   pyrosetta.py84840%15–17, 19–21, 23, 25, 27–32, 39, 42, 64–70, 73–74, 77–79, 82–83, 89, 102, 105–106, 108, 110, 112, 114, 116, 137–139, 144–145, 150–152, 155–158, 160, 162, 176–177, 183, 185, 187–188, 190–192, 194, 200, 202, 212, 214, 232–235, 238–241, 243, 247–249, 251, 256, 262, 264, 270
tools/alf_tools/models/esm_utils
   __init__.py00100% 
   config.py58198%143
   loss.py57885%67, 101, 119, 123–125, 128–129
   scoring_function.py38976%81, 91, 100, 115–119, 123
tools/alf_tools/models/utils
   __init__.py60100% 
   botorch_model_wrapper.py81396%59, 179, 244
   botorch_utils.py87495%73–75, 105
   config_utils.py27292%67, 72
   data_utils.py230100% 
   sequence_utils.py240100% 
   torch_utils.py50100% 
tools/alf_tools/models/utils/mlip_utils
   __init__.py30100% 
   config.py210100% 
   data.py90594%85, 157, 250, 257, 312
   model_registry.py13192%55
tools/alf_tools/optimizer/acquisition_functions
   __init__.py80100% 
   botorch_acquisition.py127695%290, 307, 316, 331, 410, 413
   botorch_samplers.py21385%88, 90, 117
   core_set.py46393%41–42, 95
   expected_improvement.py181327%46–52, 54, 56–58, 60, 64
   greedy.py60100% 
   thompson_sampling.py180100% 
   ucb.py13192%56
tools/alf_tools/utils
   __init__.py20100% 
   constants.py50100% 
TOTAL329029491% 

Tests Skipped Failures Errors Time
840 4 💤 0 ❌ 0 🔥 1m 28s ⏱️

chrbrunk
chrbrunk reviewed Jul 10, 2026
View reviewed changes
Comment thread .pre-commit-config.yaml Outdated
olive005 added 6 commits July 10, 2026 13:42
@olive005
chore: rm commented code
a2b5288 
Co-Authored-By: olive005 <o.gallup@instadeep.com>
@olive005
feat: update alf cover with gradient
4c4483c 
Co-Authored-By: olive005 <o.gallup@instadeep.com>
@olive005
chore: rm unused dependencies
86b6287 
Co-Authored-By: olive005 <o.gallup@instadeep.com>
@olive005
style: align python requires statements
79e133e 
Co-Authored-By: olive005 <o.gallup@instadeep.com>
@olive005
style: rm stale TODOs
04b342b 
Co-Authored-By: olive005 <o.gallup@instadeep.com>
@olive005
style: larger github cover
7d79a99 
Co-Authored-By: olive005 <o.gallup@instadeep.com>
@CLAassistant

CLAassistant commented Jul 10, 2026 •
edited
Loading

Copy link
Copy Markdown

CLA assistant check
All committers have signed the CLA.

chrbrunk
chrbrunk reviewed Jul 10, 2026
View reviewed changes
Comment thread .github/workflows/release.yml Outdated
@olive005
fix: update ci runner to ubuntu
71d114b 
Co-Authored-By: olive005 <o.gallup@instadeep.com>
chrbrunk
chrbrunk approved these changes Jul 10, 2026
View reviewed changes

@chrbrunk chrbrunk left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good to me!

@olive005
olive005 merged commit 3e25740 into main Jul 10, 2026
7 checks passed
@olive005
olive005 deleted the develop branch July 10, 2026 14:36
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@chrbrunk chrbrunk chrbrunk approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@olive005 @CLAassistant @chrbrunk