Module mapping coverage: stdlib, numpy, sklearn, pytorch

[noahgift]

Problem

Many examples fail because Python modules aren't mapped to Rust equivalents. Currently depyler emits module.function() as-is, causing E0425 (undefined) errors.

Current Module Support (partial)

Python	Rust	Status
argparse	clap	✅ Working
datetime	chrono	⚠️ Partial
pathlib	std::path	⚠️ Partial
json	serde_json	⚠️ Partial
typing	Native types	✅ Working

Modules Needed

Python Standard Library (Priority 1)

Module	Rust Crate	Examples Using
os	std::fs, std::env	15+
sys	std::env	10+
re	regex	8+
collections	std::collections	12+
itertools	itertools	6+
functools	Native closures	5+
io	std::io	10+
subprocess	std::process	4+
asyncio	tokio	5+
threading	std::thread	3+
zipfile	zip	2+
csv	csv	4+
hashlib	sha2, md5	3+
base64	base64	2+
struct	byteorder	2+
socket	std::net	2+
logging	log, tracing	5+
unittest	(skip/stub)	N/A
tempfile	tempfile	3+
shutil	fs_extra	3+
glob	glob	2+
pickle	serde + bincode	2+

NumPy (Priority 2)

Function	Rust Equivalent	Notes
np.array	Vec or ndarray	Type inference needed
np.zeros/ones	vec![0.0; n]	Shape handling
np.sum/mean/std	Iterator methods	Aggregate functions
np.dot	Manual or nalgebra	Matrix ops
np.sqrt/exp/log	f64 methods	Math functions
np.reshape	Manual indexing	Shape manipulation
np.concatenate	Vec::extend	Array joining
np.where	Iterator::filter	Conditional
Broadcasting	Manual loops	Complex

Scikit-learn (Priority 3)

Class	Rust Equivalent	Notes
train_test_split	Manual split	Simple
StandardScaler	Manual normalize	Stateful
LinearRegression	linfa	Full crate
KMeans	linfa-clustering	Full crate
metrics.*	Manual impl	accuracy, f1, etc
Pipeline	Trait chain	Design needed

PyTorch (Priority 4)

Class	Rust Equivalent	Notes
torch.Tensor	tch::Tensor	tch-rs bindings
nn.Module	Trait	Custom trait
nn.Linear	tch layers	Direct mapping
optim.Adam	Manual or tch	Optimizer
autograd	tch autograd	Complex

Proposed Implementation

Option A: Stub Generation

// Generated stub for unsupported module
mod zipfile {
    pub struct ZipFile;
    impl ZipFile {
        pub fn new(_: &str, _: &str) -> Self { todo!("zipfile not implemented") }
    }
}

Option B: Crate Mapping

# depyler.toml
[module_mappings]
"numpy" = { crate = "ndarray", features = ["blas"] }
"sklearn" = { crate = "linfa", features = ["all"] }
"torch" = { crate = "tch", features = [] }

Option C: Feature Flags

depyler transpile --features numpy,sklearn input.py

Acceptance Criteria

1. Document all stdlib modules used in examples
2. Implement top 10 stdlib modules
3. Add numpy array basics (array, zeros, sum, mean)
4. Add sklearn train_test_split
5. Emit helpful error for unmapped modules

Related

• DEPYLER-0424: zipfile module tracking
• Issue Examples needed to reach 100% depyler compilation rate #4: Examples needed for 100% compilation