Modern AlChemy – Current Issues (fontanagen branch)

How to Run

• Build: cargo build
• Run default simulation: cargo run
• Fontana generator demo: cargo run -- --config-file /path/to/fontana_config.json
• Dump default config: cargo run -- --dump-config
• Dump default config with CL: cargo run -- --config-file /tmp/fontana_cfg.json --generate 3
• Run experiments: cargo run -- --experiment <name> (not sure how to run this part)

Currently Resolved

Fontana generator looks at configuration

• src/generators.rs: FontanaGen::from_config now uses min_depth, max_depth, and free_variable_probability, clamps probability ranges, checks depth bounds, and removed the unused local RNG binding.
• Added a defensive helpers to avoid divide-by-zero (max_depth guard), empty ranges when depth is zero, and runaway probabilities (clamped per depth). Generation can now use the configured n_max_free_vars safely.
• CLI --generate now obeys the active generator in the loaded config, so Fontana samples can be printed without editing source.

Previous issue:

• src/generators.rs:202-217: min_depth and free_prob are hard-coded to 0, ignoring config::FontanaGen.min_depth and .free_variable_probability. (but it seems like in config.rs it matches the same values? do we hard code or use that?)
• src/generators.rs:197-200: the local rng binding is never used; we instantiate the struct with ChaCha8Rng::from_seed(seed) directly, so this variable only triggers warnings.
• src/generators.rs:197-200: division by cfg.max_depth - 1 won't work if max_depth <= 1.
• src/generators.rs:236-279: p_abs/p_app can exceed 1.0, meaning leaf nodes might never be produced before hitting maximum depth; when depth == 0, rng.gen_range(1..=depth) is an empty range.
• src/main.rs:156: --generate always uses BTreeGen. Need to do for FontanaGen

Warning cleanup

• Removed unused helper functions and tightened imports in src/experiments/magic_test_function.rs; wrapped tests in #[cfg(test)] to avoid dev-build noise.
• Added documentation for config::FontanaGen, so the #[warn(missing_docs)] lint is satisfied.
• cargo build and cargo test now run warning-free.

Remaining Known Issues

Postfix standardization unimplemented

• src/generators.rs:134-136: postfix_standardize calls unimplemented!.
• Any config selecting Standardization::Postfix crashes.
• Need to add the transformation or disable the option in configs/CLI until ready.

Collision metrics mislabelled

• src/lambda/recursive.rs:227-232: tuples are pushed as (expr, size, reductions) but collected as reductions → t.1, sizes → t.2, swapping the data.
• **Does this imapct antyhing? does it affect the analytics and experiment outputs report incorrect reduction counts vs sizes.

Jacard metric incorrect

• src/analysis.rs:57-68: calculates intersection / (|A| + |B|)
• should be divded by A U B

Documentation

• scripts/discovery.sh:15 (and similar files): cd ~/cwd/functional-supercollider, which doesn’t exist here.

Interactive Dashboard

Can be found here: https://github.com/mathis-group/Alchemy-Dashboard

Steps to run updated

0) Pick & activate the EXACT Python you will use (avoid shim confusion)

(choose one)

— pyenv:

pyenv shell 3.11.9

— OR virtualenv:

python3.11 -m venv .venv && source .venv/bin/activate

1) Show which Python we’re about to use

python -c 'import sys,platform; print(sys.executable); print(platform.python_version(), platform.machine())'

2) Clean previous artifacts

cargo clean python -m pip uninstall -y alchemy || true rm -rf target

3) Build & install the package editable via the PEP 517 backend (maturin)

python -m pip install -e .

4) Sanity-check the import (this uses the same interpreter as above)

python - <<'PY' import alchemy, sys print("OK :", alchemy.file) print("Py :", sys.executable) PY