Skip to content

examples.md

Latest commit

 

History

History
95 lines (67 loc) · 3.12 KB

examples.md

File metadata and controls

95 lines (67 loc) · 3.12 KB
Raw

Examples

Every snippet here is self-contained — algo2code has no runtime dependencies, so these run anywhere the package is importable.

!!! tip "Run them with uv" Inside the workspace, prefix commands with uv run (e.g. uv run python ex.py) after uv sync --all-packages --all-groups --all-extras.

Transpile a library algorithm

The fastest path is the convenience wrappers — each returns generated Taichi source as a string:

from algo2code.library.radial_return_j2 import transpile_radial_return_j2

code = transpile_radial_return_j2(backend="taichi")
print(code)        # deterministic Taichi-compatible Python source

Transpile from the verbatim LaTeX

Equivalent to the wrapper, but shows the LaTeX-is-the-source-of-truth path explicitly:

from algo2code import transpile
from algo2code.library.radial_return_j2 import RADIAL_RETURN_J2_LATEX

code = transpile(RADIAL_RETURN_J2_LATEX, backend="taichi")

The full J2 family

All three return-map variants follow the same shape — only the constant and wrapper change:

from algo2code.library.radial_return_j2 import RADIAL_RETURN_J2_LATEX
from algo2code.library.radial_return_j2_kinematic import RADIAL_RETURN_J2_KINEMATIC_LATEX
from algo2code.library.radial_return_j2_mixed import RADIAL_RETURN_J2_MIXED_LATEX
from algo2code import transpile

for name, latex in [
    ("isotropic power-law", RADIAL_RETURN_J2_LATEX),
    ("linear kinematic",    RADIAL_RETURN_J2_KINEMATIC_LATEX),
    ("mixed hardening",     RADIAL_RETURN_J2_MIXED_LATEX),
]:
    code = transpile(latex, backend="taichi")
    print(f"{name}: {len(code)} chars of generated source")

These scalar return-maps are the inner loops of the J2 plasticity models in mechdsl-core — see Usage → how the J2 family is wired.

Transpile the PCG solver

from algo2code import transpile, PCG_ALGORITHM_LATEX

code = transpile(PCG_ALGORITHM_LATEX, backend="taichi")

The generated PCG is the single source of truth behind mechdsl-core's Algo2CodePCGSolver, selectable in the Newton driver via select_linear_solver("generated") — see Usage → the solver seam.

Exec generated code into a callable

transpile returns source text. To get a function you can call, exec it. The emitted function name matches the algorithm name:

from algo2code import transpile
from algo2code.library.radial_return_j2 import RADIAL_RETURN_J2_LATEX

code = transpile(RADIAL_RETURN_J2_LATEX, backend="taichi")

ns: dict = {}
exec(compile(code, "<algo2code>", "exec"), ns)
radial_return_j2 = ns["radial_return_j2"]   # the scalar return-map, now callable

Inspect the parsed AST

The intermediate stages are exported, so you can look at what the parser and type inference produced before codegen:

from algo2code import parse_algorithm, infer_types
from algo2code.library.radial_return_j2 import RADIAL_RETURN_J2_LATEX

algo = parse_algorithm(RADIAL_RETURN_J2_LATEX)   # Algorithm AST
infer_types(algo)                                # annotate scalar/array types in place
print(algo.name)