ml_switcheroo.testing.fuzzer.generators¶

Primitive Generation Logic for the Fuzzer.

This module contains functions to generate random scalars, NumPy arrays, and dummy callables respecting specified constraints (min, max, dtype).

Feature Update: Added broadcasting logic generator.

Functions¶

`generate_scalar_int`(→ int)	Generates a random integer within constrained bounds.
`generate_scalar_float`(→ float)	Generates a random float within constrained bounds.
`generate_array`(→ numpy.ndarray)	Generates a random NumPy array bounded by constraints.
`get_random_shape`(→ Tuple[int, Ellipsis])	Selects a random rank (1-4) and random dimensions (2-5).
`make_broadcastable_shape`(→ Tuple[int, Ellipsis])	Derives a shape that is broadcast-compatible with the base_shape.
`generate_fake_callable`(→ Any)	Generates a dummy function (identity) for functional ops.

Module Contents¶

ml_switcheroo.testing.fuzzer.generators.generate_scalar_int(constraints: Dict[str, Any]) → int¶

Generates a random integer within constrained bounds.

Parameters:: constraints (Dict[str, Any]) – Dictionary containing optional ‘min’ and ‘max’.
Returns:: A random integer.
Return type:: int

ml_switcheroo.testing.fuzzer.generators.generate_scalar_float(constraints: Dict[str, Any]) → float¶

Generates a random float within constrained bounds.

Parameters:: constraints (Dict[str, Any]) – Dictionary containing optional ‘min’ and ‘max’.
Returns:: A random float.
Return type:: float

ml_switcheroo.testing.fuzzer.generators.generate_array(type_lbl: str, shape: Tuple[int, Ellipsis], constraints: Dict[str, Any]) → numpy.ndarray¶

Generates a random NumPy array bounded by constraints.

Parameters:

type_lbl (str) – General type category (‘float’, ‘int’, ‘bool’).
shape (Tuple[int, ...]) – The shape of the array.
constraints (Dict[str, Any]) – Constraints like ‘min’, ‘max’, ‘dtype’.

Returns:

The generated array.

Return type:

np.ndarray

ml_switcheroo.testing.fuzzer.generators.get_random_shape(seed_shape: Tuple[int, Ellipsis] | None = None) → Tuple[int, Ellipsis]¶

Selects a random rank (1-4) and random dimensions (2-5).

Parameters:: seed_shape (Optional[Tuple]) – Optional fixed shape to return.
Returns:: The random or seeded shape.
Return type:: Tuple[int, …]

ml_switcheroo.testing.fuzzer.generators.make_broadcastable_shape(base_shape: Tuple[int, Ellipsis], salt: int = 0) → Tuple[int, Ellipsis]¶

Derives a shape that is broadcast-compatible with the base_shape.

For each dimension, there is a probability it becomes 1 (broadcasting dimension). The ‘salt’ ensures arguments don’t all degenerate to 1s in the same way, creating varied valid broadcasting patterns (e.g. (A, 1, C) vs (1, B, C)).

Parameters:

base_shape – The target accumulated shape.
salt – Integer to vary the random choice (arg index).

Returns:

A new shape tuple.

ml_switcheroo.testing.fuzzer.generators.generate_fake_callable(constraints: Dict[str, Any] = None) → Any¶: Generates a dummy function (identity) for functional ops.