core

This module provides the core class Solver, which loads a mode and finds Rubik's Cube solutions using a beam search algorithm.

Class:

• Solver: A class for managing Rubik's Cube configuration, solving model, and search function.

Solver

class Solver()

A solver class for managing environment, model, and search configurations.

This class orchestrates the cube environment, the neural network model, and the beam search algorithm to find solutions.

Methods:

• load: Load the solver model and optimize it for CPU or GPU.
• __call__ (or solve): Set up the cube state and find solutions using beam search.
• benchmark: Evaluate the solver's search efficiency.
• evaluate_temporal_performance: (Deprecated) Evaluate solution length vs. time.

load

def load(
    model_id: str = dict(cpu="small").get(device.type, "large"),
    prefer_gpu: bool = True,
    quantize_on_cpu: bool = True,
    jit_mode: bool = False,
    cache_dir: str = os.path.expanduser("~/.cache/alphacube")
)

Load the Rubik's Cube solver model and optimize it for CPU or GPU.

Arguments:

• model_id str - Identifier for the model variant to load ("small", "base", or "large").
• prefer_gpu bool - Whether to prefer GPU if available.
• quantize_on_cpu bool - Whether to quantize the model for CPU optimization.
• jit_mode bool - Whether to enable JIT mode for potentially faster execution.
• cache_dir str - Directory to cache the model files.

Returns:

• None

__call__

def __call__(
    scramble,
    format="moves",
    allow_wide=True,
    beam_width: int = 1024,
    extra_depths: int = 0,
    ergonomic_bias: dict | None = None
)

Set up the cube state and find solutions using beam search.

Arguments:

• scramble list or str - A sequence of moves or a sticker representation of the initial cube state.
• format str - Input format of the scramble: "moves" or "stickers".
• allow_wide bool - Whether to allow wide moves in the search space.
• Note - This is often controlled automatically by ergonomic_bias.
• beam_width int - The beam width for the search algorithm.
• extra_depths int - Number of extra depths to search after finding a solution.
• ergonomic_bias dict, optional - A dictionary to bias the search towards certain moves based on ergonomic preference.

Returns:

• dict | None: A dictionary containing the solution(s) and search metadata, or None if no solution is found.

benchmark

A shortcut method alphacube.solver.benchmark(...) functionally equivalent to alphacube._evaluator.benchmark(alphacube.solver, ...). evaluate_search_efficiency redirects to this method for backward compatibility.

evaluate_temporal_performance

Deprecated. A shortcut alphacube._evaluator.evaluate_temporal_performance