from __future__ import annotations
import random
from mtdata.forecast import tune
def test_default_search_space_modes():
multi = tune.default_search_space(methods=["theta", "fourier_ols"])
assert "_shared" in multi
assert "theta" in multi and "seasonality" in multi["theta"]
assert "fourier_ols" in multi and "K" in multi["fourier_ols"]
single_known = tune.default_search_space(method="theta")
assert "seasonality" in single_known
single_unknown = tune.default_search_space(method="unknown")
assert single_unknown == {"seasonality": {"type": "int", "min": 8, "max": 48}}
none_given = tune.default_search_space()
assert "_shared" in none_given
assert "theta" in none_given
def test_sample_and_mutate_param_helpers():
rng = random.Random(7)
assert tune._sample_param({"type": "categorical", "choices": ["a", "b"]}, rng) in {"a", "b"}
assert tune._sample_param({"type": "categorical", "choices": []}, rng) is None
assert tune._sample_param({"type": "int", "min": 5, "max": 3}, rng) in {3, 4, 5}
assert isinstance(tune._sample_param({"type": "float", "min": 0.1, "max": 0.2, "log": True}, rng), float)
assert tune._mutate_value("a", {"type": "categorical", "choices": ["a", "b"]}, rng) == "b"
assert tune._mutate_value("a", {"type": "categorical", "choices": ["a"]}, rng) == "a"
assert tune._mutate_value(5, {"type": "int", "min": 0, "max": 10}, rng) >= 0
assert 0.0 <= tune._mutate_value(0.5, {"type": "float", "min": 0.0, "max": 1.0}, rng) <= 1.0
def test_crossover_for_method_blends_and_fills_none():
rng = random.Random(13)
a = {"x": 1.0, "cat": "a", "i": None}
b = {"x": 3.0, "cat": "b", "i": None}
spaces = {
"x": {"type": "float", "min": 0.0, "max": 10.0},
"cat": {"type": "categorical", "choices": ["a", "b"]},
"i": {"type": "int", "min": 1, "max": 2},
}
child = tune._crossover_for_method(a, b, spaces, rng)
assert child["x"] == 2.0
assert child["cat"] in {"a", "b"}
assert child["i"] in {1, 2}
def test_eval_candidate_handles_method_selection_and_failures(monkeypatch):
def fake_backtest(**kwargs):
m = kwargs["methods"][0]
if m == "bad":
return {"results": {m: {"success": False}}}
if m == "missing_metric":
return {"results": {m: {"success": True, "avg_mae": 2.5}}}
return {"results": {m: {"success": True, "avg_rmse": 1.2}}}
monkeypatch.setattr(tune, "_forecast_backtest", fake_backtest)
score, result = tune._eval_candidate(
symbol="EURUSD",
timeframe="H1",
method="theta",
horizon=2,
steps=2,
spacing=1,
candidate_params={},
metric="avg_rmse",
mode="min",
)
assert score == 1.2
assert result["_sel_method"] == "theta"
score, _ = tune._eval_candidate(
symbol="EURUSD",
timeframe="H1",
method="theta",
horizon=2,
steps=2,
spacing=1,
candidate_params={"method": "bad"},
metric="avg_rmse",
mode="min",
)
assert score == float("inf")
score, result = tune._eval_candidate(
symbol="EURUSD",
timeframe="H1",
method=None,
horizon=2,
steps=2,
spacing=1,
candidate_params={"method": "missing_metric"},
metric="avg_rmse",
mode="max",
)
assert score == -2.5
assert result["_sel_method"] == "missing_metric"
score, result = tune._eval_candidate(
symbol="EURUSD",
timeframe="H1",
method=None,
horizon=2,
steps=2,
spacing=1,
candidate_params={},
metric="avg_rmse",
mode="min",
)
assert score == float("inf")
assert result["error"] == "No method provided"
def test_genetic_search_method_scoped_and_flat_spaces(monkeypatch):
def fake_eval_candidate(**kwargs):
cand = kwargs["candidate_params"]
m = cand.get("method") or kwargs.get("method") or "theta"
score_val = float(cand.get("x", 1.0))
return (
score_val if kwargs["mode"] == "min" else -score_val,
{"_sel_method": m, "results": {m: {"horizon": kwargs["horizon"], "success": True}}},
)
monkeypatch.setattr(tune, "_eval_candidate", fake_eval_candidate)
method_scoped_space = {
"_shared": {"x": {"type": "float", "min": 0.1, "max": 1.0}},
"theta": {"k": {"type": "int", "min": 1, "max": 3}},
"naive": {"k": {"type": "int", "min": 4, "max": 6}},
}
out = tune.genetic_search_forecast_params(
symbol="EURUSD",
timeframe="H1",
method=None,
methods=["theta", "naive"],
horizon=3,
steps=2,
spacing=1,
search_space=method_scoped_space,
population=4,
generations=2,
seed=11,
)
assert out["success"] is True
assert out["history_count"] == 8
assert out["best_method"] in {"theta", "naive"}
assert "best_result_summary" in out
assert len(out["history_tail"]) <= 50
flat_space = {
"method": {"type": "categorical", "choices": ["theta", "naive"]},
"x": {"type": "float", "min": 0.2, "max": 0.9},
}
out = tune.genetic_search_forecast_params(
symbol="EURUSD",
timeframe="H1",
method=None,
methods=None,
horizon=2,
steps=2,
spacing=1,
search_space=flat_space,
mode="max",
population=3,
generations=2,
seed=17,
)
assert out["success"] is True
assert out["mode"] == "max"
assert out["history_count"] == 6
