rcf3 0.2.0

import math
import pickle
from collections.abc import Callable
from copy import deepcopy

import pytest
from hypothesis import given, settings
from hypothesis import strategies as st

from rcf3 import Forest

DEFAULT_NUM_TREES = 40
DEFAULT_CAPACITY = 64
DEFAULT_OUTPUT_AFTER = 1


FINITE_F32 = st.floats(
    min_value=-1_000.0,
    max_value=1_000.0,
    allow_nan=False,
    allow_infinity=False,
    width=32,
)


def vector_strategy(dim: int) -> st.SearchStrategy[list[float]]:
    return st.lists(FINITE_F32, min_size=dim, max_size=dim)


def make_forest(  # noqa: PLR0913
    *,
    input_dim: int,
    seed: int,
    shingle_size: int = 1,
    internal_shingling: bool = True,
    num_trees: int = DEFAULT_NUM_TREES,
    capacity: int = DEFAULT_CAPACITY,
    output_after: int = DEFAULT_OUTPUT_AFTER,
) -> Forest:
    """Create a Forest test fixture with deterministic defaults."""
    return Forest(
        input_dim=input_dim,
        shingle_size=shingle_size,
        num_trees=num_trees,
        capacity=capacity,
        output_after=output_after,
        internal_shingling=internal_shingling,
        seed=seed,
    )


def warm_forest(
    *,
    input_dim: int,
    shingle_size: int = 1,
    internal_shingling: bool = True,
    seed: int = 7,
    updates: int = 64,
) -> Forest:
    """Create a forest and warm it with synthetic observations."""
    forest = make_forest(
        input_dim=input_dim,
        seed=seed,
        shingle_size=shingle_size,
        internal_shingling=internal_shingling,
    )
    obs_dim = input_dim if internal_shingling else input_dim * shingle_size
    base = [0.125] * obs_dim
    for i in range(updates):
        point = [x + (i % 5) * 0.01 for x in base]
        forest.update(point)
    return forest


@settings(max_examples=30, deadline=None)
@given(
    input_dim=st.integers(min_value=1, max_value=6),
    stream_len=st.integers(min_value=24, max_value=72),
    seed=st.integers(min_value=0, max_value=2**32 - 1),
    data=st.data(),
)
def test_seeded_stream_is_deterministic(
    input_dim: int,
    stream_len: int,
    seed: int,
    data: st.DataObject,
) -> None:
    stream = [
        data.draw(vector_strategy(input_dim), label=f"p{i}") for i in range(stream_len)
    ]
    queries = [data.draw(vector_strategy(input_dim), label=f"q{i}") for i in range(6)]

    f1 = make_forest(input_dim=input_dim, seed=seed)
    f2 = make_forest(input_dim=input_dim, seed=seed)

    for point in stream:
        f1.update(point)
        f2.update(point)

    assert f1.entries_seen() == f2.entries_seen()
    for q in queries:
        assert f1.score(q) == pytest.approx(f2.score(q), abs=1e-12)


@settings(max_examples=25, deadline=None)
@given(
    input_dim=st.integers(min_value=1, max_value=5),
    shingle_size=st.integers(min_value=1, max_value=4),
    capacity=st.integers(min_value=8, max_value=128),
    internal_shingling=st.booleans(),
    output_after=st.integers(min_value=0, max_value=20),
)
def test_is_ready_threshold_contract(
    input_dim: int,
    shingle_size: int,
    capacity: int,
    internal_shingling: bool,
    output_after: int,
) -> None:
    forest = make_forest(
        input_dim=input_dim,
        seed=19,
        shingle_size=shingle_size,
        num_trees=30,
        capacity=capacity,
        output_after=output_after,
        internal_shingling=internal_shingling,
    )

    effective_output_after = (1 + capacity // 4) if output_after == 0 else output_after
    needed = effective_output_after + (shingle_size - 1 if internal_shingling else 0)
    obs_dim = input_dim if internal_shingling else input_dim * shingle_size
    obs = [0.0] * obs_dim

    for _ in range(needed):
        forest.update(obs)
        assert not forest.is_ready()

    forest.update(obs)
    assert forest.is_ready()


class TestRoundTrip:
    """Round-trip serialization tests preserving state and scoring."""

    def _inner(
        self,
        input_dim: int,
        seed: int,
        data: st.DataObject,
        roundtrip: Callable[[Forest], Forest],
    ) -> None:
        """Assert that a round-trip transform preserves entries and score."""
        forest = make_forest(input_dim=input_dim, seed=seed)
        points = [
            data.draw(vector_strategy(input_dim), label=f"u{i}") for i in range(48)
        ]
        query = data.draw(vector_strategy(input_dim), label="query")

        for point in points:
            forest.update(point)

        before = forest.score(query)
        restored = roundtrip(forest)

        assert restored.entries_seen() == forest.entries_seen()
        assert restored.score(query) == pytest.approx(before, abs=1e-12)

    @settings(max_examples=20, deadline=None)
    @given(
        input_dim=st.integers(min_value=1, max_value=6),
        seed=st.integers(min_value=0, max_value=2**32 - 1),
        data=st.data(),
    )
    def test_json_roundtrip_preserves_state_and_scores(
        self,
        input_dim: int,
        seed: int,
        data: st.DataObject,
    ) -> None:
        self._inner(
            input_dim=input_dim,
            seed=seed,
            data=data,
            roundtrip=lambda f: Forest.from_json(f.to_json()),
        )

    @settings(max_examples=20, deadline=None)
    @given(
        input_dim=st.integers(min_value=1, max_value=6),
        seed=st.integers(min_value=0, max_value=2**32 - 1),
        protocol=st.integers(
            min_value=pickle.DEFAULT_PROTOCOL, max_value=pickle.HIGHEST_PROTOCOL
        ),
        data=st.data(),
    )
    def test_pickle_roundtrip_preserves_state_and_scores(
        self,
        input_dim: int,
        seed: int,
        protocol: int,
        data: st.DataObject,
    ) -> None:
        self._inner(
            input_dim=input_dim,
            seed=seed,
            data=data,
            roundtrip=lambda f: pickle.loads(pickle.dumps(f, protocol=protocol)),  # noqa: S301
        )

    @settings(max_examples=20, deadline=None)
    @given(
        input_dim=st.integers(min_value=1, max_value=6),
        seed=st.integers(min_value=0, max_value=2**32 - 1),
        data=st.data(),
    )
    def test_deepcopy_preserves_state_and_scores(
        self,
        input_dim: int,
        seed: int,
        data: st.DataObject,
    ) -> None:
        self._inner(
            input_dim=input_dim,
            seed=seed,
            data=data,
            roundtrip=deepcopy,
        )


@settings(max_examples=20, deadline=None)
@given(
    input_dim=st.integers(min_value=2, max_value=6),
    top_k=st.integers(min_value=1, max_value=12),
    percentile=st.integers(min_value=0, max_value=100),
    data=st.data(),
)
def test_near_neighbors_are_bounded_and_sorted(
    input_dim: int,
    top_k: int,
    percentile: int,
    data: st.DataObject,
) -> None:
    forest = make_forest(input_dim=input_dim, seed=23, num_trees=50)
    updates = [data.draw(vector_strategy(input_dim), label=f"u{i}") for i in range(56)]
    query = data.draw(vector_strategy(input_dim), label="query")

    for point in updates:
        forest.update(point)

    results = forest.near_neighbors(query, top_k=top_k, percentile=percentile)
    assert len(results) <= top_k
    assert all(len(item["point"]) == input_dim for item in results)

    distances = [item["distance"] for item in results]
    assert distances == sorted(distances)


@settings(max_examples=25, deadline=None)
@given(
    input_dim=st.integers(min_value=2, max_value=8),
    data=st.data(),
)
def test_impute_deterministic_with_centrality_one_and_preserves_observed(
    input_dim: int,
    data: st.DataObject,
) -> None:
    forest = warm_forest(input_dim=input_dim, seed=29)
    query = data.draw(vector_strategy(input_dim), label="query")
    missing = sorted(
        data.draw(
            st.sets(
                st.integers(min_value=0, max_value=input_dim - 1),
                min_size=1,
                max_size=input_dim - 1,
            ),
            label="missing",
        )
    )

    out1 = forest.impute(query, missing, centrality=1.0)
    out2 = forest.impute(query, missing, centrality=1.0)
    out3 = forest.impute(query, missing, centrality=1.0)

    assert len(out1) == input_dim
    assert out1 == pytest.approx(out2, abs=1e-7)
    assert out2 == pytest.approx(out3, abs=1e-7)

    missing_set = set(missing)
    for i, original in enumerate(query):
        if i not in missing_set:
            assert out1[i] == pytest.approx(original, abs=1e-6)
        assert math.isfinite(out1[i])


@settings(max_examples=15, deadline=None)
@given(
    input_dim=st.integers(min_value=1, max_value=4),
    shingle_size=st.integers(min_value=2, max_value=4),
    data=st.data(),
)
def test_extrapolate_shape_and_finite_values(
    input_dim: int,
    shingle_size: int,
    data: st.DataObject,
) -> None:
    look_ahead = data.draw(
        st.integers(min_value=0, max_value=shingle_size), label="look_ahead"
    )
    forest = warm_forest(
        input_dim=input_dim,
        shingle_size=shingle_size,
        internal_shingling=True,
        seed=31,
        updates=96,
    )
    out = forest.extrapolate(look_ahead)

    assert len(out) == look_ahead * input_dim
    assert all(math.isfinite(x) for x in out)


@settings(max_examples=10, deadline=None)
@given(
    input_dim=st.integers(min_value=1, max_value=5),
    look_ahead=st.integers(min_value=1, max_value=5),
)
def test_extrapolate_requires_internal_shingling(
    input_dim: int, look_ahead: int
) -> None:
    forest = warm_forest(
        input_dim=input_dim,
        shingle_size=2,
        internal_shingling=False,
        seed=37,
        updates=96,
    )
    with pytest.raises(ValueError, match="internal_shingling"):
        forest.extrapolate(look_ahead)


@settings(max_examples=10, deadline=None)
@given(
    input_dim=st.integers(min_value=1, max_value=5),
    look_ahead=st.integers(min_value=1, max_value=5),
)
def test_extrapolate_requires_shingle_size_gt_one(
    input_dim: int, look_ahead: int
) -> None:
    forest = warm_forest(
        input_dim=input_dim,
        shingle_size=1,
        internal_shingling=True,
        seed=41,
        updates=96,
    )
    with pytest.raises(ValueError, match="shingle_size"):
        forest.extrapolate(look_ahead)


@settings(max_examples=12, deadline=None)
@given(
    input_dim=st.integers(min_value=1, max_value=5),
    shingle_size=st.integers(min_value=2, max_value=5),
    extra=st.integers(min_value=1, max_value=5),
)
def test_extrapolate_rejects_look_ahead_beyond_shingle_window(
    input_dim: int,
    shingle_size: int,
    extra: int,
) -> None:
    forest = warm_forest(
        input_dim=input_dim,
        shingle_size=shingle_size,
        internal_shingling=True,
        seed=43,
        updates=96,
    )
    with pytest.raises(ValueError, match="look_ahead"):
        forest.extrapolate(shingle_size + extra)


@settings(max_examples=15, deadline=None)
@given(input_dim=st.integers(min_value=2, max_value=8), data=st.data())
def test_dimension_mismatch_raises_value_error(
    input_dim: int, data: st.DataObject
) -> None:
    forest = make_forest(input_dim=input_dim, seed=11)
    bad_point = data.draw(vector_strategy(input_dim + 1), label="bad")

    with pytest.raises(ValueError, match="dimension mismatch"):
        forest.update(bad_point)


def test_impute_not_ready_raises_runtime_error() -> None:
    forest = make_forest(input_dim=4, seed=3)

    assert not forest.is_ready()
    with pytest.raises(RuntimeError):
        forest.impute([0.1, 0.2, 0.3, 0.4], [1], centrality=1.0)