ml_switcheroo.core.discovery¶

Core Discovery mechanism for Auto-Inference of APIs.

This module provides the SimulatedReflection engine, which enables the system to dynamically locate API endpoints in a target framework based on an abstract operation name. It is used by the define CLI command to resolve api: “infer” placeholders in ODL (Operation Definition Language) files.

Attributes¶

logger

Classes¶

SimulatedReflection

Scans a target framework to infer the API path for a given operation name.

Module Contents¶

ml_switcheroo.core.discovery.logger¶

class ml_switcheroo.core.discovery.SimulatedReflection(framework: str)¶

Scans a target framework to infer the API path for a given operation name.

It utilizes the FrameworkAdapter.search_modules to limit the search space and applies both exact normalization matching and fuzzy string matching to find the best candidate.

framework¶

adapter¶

discover(op_name: str) → str | None¶

Attempts to locate the fully qualified API path for an operation.

Strategy: 1. Normalization: Converts input and candidates to lowercase, stripping underscores. 2. Exact Match: Iterates through search modules. If a normalized match is found, returns immediately. 3. Fuzzy Match: If no exact match is found, collects all candidates and uses difflib

to find the closest string match.

Parameters:: op_name (str) – The abstract operation name (e.g. “LogSoftmax”, “abs”).
Returns:: The fully qualified API path (e.g. “torch.nn.functional.log_softmax”) or None if no confident match is found.
Return type:: Optional[str]