danielmjrr
160ed2f948
Memory-mapped tickstores, a nogil numba replay kernel validated trade-for-trade against an independent reference implementation, purged walk-forward CV, an append-only FDR registry, tick-exact labels, and online changepoint detection. 129 tests, 4 runnable examples on synthetic data, CI for linux/macos.
91 lines
2.6 KiB
Python
91 lines
2.6 KiB
Python
import numpy as np
|
|
import pytest
|
|
|
|
from tickreplay.research.registry import REQUIRED, append, load
|
|
from tickreplay.research.stats import bh_qvalues
|
|
|
|
|
|
def row(config, p_value, **over):
|
|
base = {"campaign": "c1", "symbol": "SYNTH", "config": config, "n_trades": 120,
|
|
"mean_net": 0.02, "t_stat": 2.0, "p_value": p_value}
|
|
base.update(over)
|
|
return base
|
|
|
|
|
|
def test_append_rejects_rows_missing_required_columns(tmp_path):
|
|
path = tmp_path / "registry.csv"
|
|
incomplete = row("a", 0.01)
|
|
del incomplete["p_value"]
|
|
|
|
with pytest.raises(ValueError, match="p_value"):
|
|
append(path, [incomplete])
|
|
|
|
assert not path.exists()
|
|
|
|
|
|
def test_append_validates_every_row_before_writing_any(tmp_path):
|
|
path = tmp_path / "registry.csv"
|
|
bad = row("b", 0.01)
|
|
del bad["n_trades"]
|
|
|
|
with pytest.raises(ValueError):
|
|
append(path, [row("a", 0.01), bad])
|
|
|
|
assert not path.exists()
|
|
|
|
|
|
@pytest.mark.parametrize("column", REQUIRED)
|
|
def test_every_required_column_is_enforced(tmp_path, column):
|
|
path = tmp_path / f"registry_{column}.csv"
|
|
incomplete = row("a", 0.01)
|
|
del incomplete[column]
|
|
|
|
with pytest.raises(ValueError, match=column):
|
|
append(path, [incomplete])
|
|
|
|
|
|
def test_append_twice_then_load_returns_all_rows_with_qvalues(tmp_path):
|
|
path = tmp_path / "nested" / "registry.csv"
|
|
|
|
append(path, [row("a", 0.01), row("b", 0.50)])
|
|
append(path, [row("c", 0.60), row("d", 0.70)])
|
|
|
|
df = load(path)
|
|
|
|
assert len(df) == 4
|
|
assert list(df["config"]) == ["a", "b", "c", "d"]
|
|
assert "q_value" in df.columns
|
|
assert "ts" in df.columns
|
|
assert set(REQUIRED) <= set(df.columns)
|
|
np.testing.assert_allclose(df["q_value"].to_numpy(),
|
|
bh_qvalues(df["p_value"].to_numpy()))
|
|
|
|
|
|
def test_qvalues_are_recomputed_cumulatively_not_stored(tmp_path):
|
|
path = tmp_path / "registry.csv"
|
|
|
|
append(path, [row("a", 0.01), row("b", 0.50)])
|
|
first = load(path)
|
|
assert first.loc[0, "q_value"] == pytest.approx(0.02)
|
|
assert first.loc[1, "q_value"] == pytest.approx(0.50)
|
|
|
|
append(path, [row("c", 0.60), row("d", 0.70)])
|
|
second = load(path)
|
|
|
|
assert second.loc[0, "q_value"] == pytest.approx(0.04)
|
|
assert second.loc[1, "q_value"] == pytest.approx(0.70)
|
|
assert second.loc[0, "q_value"] > first.loc[0, "q_value"]
|
|
assert "q_value" not in path.read_text().splitlines()[0]
|
|
|
|
|
|
def test_registry_is_append_only(tmp_path):
|
|
path = tmp_path / "registry.csv"
|
|
|
|
append(path, [row("a", 0.01)])
|
|
append(path, [row("a", 0.01)])
|
|
|
|
df = load(path)
|
|
|
|
assert len(df) == 2
|
|
assert path.read_text().count("campaign,symbol") == 1
|