Codex
879 lines
41 KiB
Python
879 lines
41 KiB
Python
from __future__ import annotations
|
|
|
|
import sys
|
|
import tempfile
|
|
import unittest
|
|
from argparse import Namespace
|
|
from pathlib import Path
|
|
|
|
import numpy as np
|
|
import pandas as pd
|
|
|
|
TRAINING_ROOT = Path(__file__).resolve().parents[1]
|
|
if str(TRAINING_ROOT) not in sys.path:
|
|
sys.path.insert(0, str(TRAINING_ROOT))
|
|
|
|
from trader_training.onnx_export import LinearHead, export_heads
|
|
from trader_training.conditional_entry_probe import probe_conditional_entry_training
|
|
from trader_training.direction_opportunity_dataset import _opportunity_labels
|
|
from trader_training.dynamic_exit_search import search_dynamic_exit_plans
|
|
from trader_training.entry_condition_pair_screen import screen_entry_condition_pairs
|
|
from trader_training.entry_feature_screen import _bucket_edges, _screen_edge_column
|
|
from trader_training.entry_mae_label_diagnostic import diagnose_entry_mae_labels
|
|
from trader_training.good_trade_structure import _side_frame, _top_fraction_metrics
|
|
from trader_training.io_utils import read_json, write_json
|
|
from trader_training.labels import ENTRY_LABEL_METHOD, _path_stats_for_group, build_entry_labels
|
|
from trader_training.nonlinear_pm_probe import _entry_side_fit_frame, _exit_metrics, _expanded_threshold_candidates
|
|
from trader_training.ofi_feature_experiment import _load_entry_dataset, l1_snapshot_diff_ofi_quote
|
|
from trader_training.promote import promote_artifact_bundle
|
|
from trader_training.replay import build_splits
|
|
from trader_training.schemas import FEATURE_ORDER, LATEST_STRESS_SPLIT, MODEL_OUTPUTS, OUTPUT_MAPPING, TRAINING_SPLITS, VALIDATION_LOCKED_SPLIT
|
|
from trader_training.training import TARGETS, _head_train_mask
|
|
from trader_training.diagnostics import _label_summary
|
|
|
|
|
|
class TrainingContractTest(unittest.TestCase):
|
|
def test_feature_order_is_v4_contract_size(self) -> None:
|
|
self.assertEqual(54, len(FEATURE_ORDER))
|
|
self.assertEqual(len(FEATURE_ORDER), len(set(FEATURE_ORDER)))
|
|
self.assertEqual("ret_1m_bps", FEATURE_ORDER[0])
|
|
self.assertEqual("book_pressure_reversal_15m", FEATURE_ORDER[-1])
|
|
|
|
def test_output_mapping_matches_model_outputs(self) -> None:
|
|
for model_name, fields in MODEL_OUTPUTS.items():
|
|
self.assertEqual(set(fields), set(OUTPUT_MAPPING[model_name]))
|
|
self.assertEqual([f"prediction[{idx}]" for idx in range(len(fields))], [OUTPUT_MAPPING[model_name][field] for field in fields])
|
|
|
|
def test_nonlinear_pm_probe_expands_low_probability_thresholds(self) -> None:
|
|
candidates = _expanded_threshold_candidates()
|
|
|
|
self.assertIn(
|
|
{
|
|
"long_open_prob": 0.2,
|
|
"short_open_prob": 0.2,
|
|
"min_entry_prob": 0.05,
|
|
"max_market_risk_prob": 0.45,
|
|
"min_expected_edge_bps": -5.0,
|
|
"min_direction_margin": 0.0,
|
|
},
|
|
candidates,
|
|
)
|
|
self.assertIn(
|
|
{
|
|
"long_open_prob": 1.01,
|
|
"short_open_prob": 0.2,
|
|
"min_entry_prob": 0.05,
|
|
"max_market_risk_prob": 0.45,
|
|
"min_expected_edge_bps": -5.0,
|
|
"min_direction_margin": 0.0,
|
|
},
|
|
candidates,
|
|
)
|
|
|
|
def test_nonlinear_entry_tree_probe_can_use_side_opportunity_rows(self) -> None:
|
|
direction = pd.DataFrame(
|
|
{
|
|
"sample_id": ["s1", "s2", "s3", "s4"],
|
|
"long_target": [1, 0, 0, 0],
|
|
"short_target": [0, 1, 0, 0],
|
|
}
|
|
)
|
|
entry = pd.DataFrame(
|
|
{
|
|
"sample_id": ["s1", "s2", "s3", "s4"],
|
|
"split_id": ["fit_inner", "fit_inner", "fit_inner", "fit_inner"],
|
|
"long_max_achievable_net_edge_bps": [45.0, 10.0, 65.0, 39.0],
|
|
"short_max_achievable_net_edge_bps": [8.0, 41.0, 15.0, 70.0],
|
|
}
|
|
)
|
|
|
|
long_frame = _entry_side_fit_frame(direction, entry, "LONG", "side_opportunity", 40.0)
|
|
short_frame = _entry_side_fit_frame(direction, entry, "SHORT", "side_opportunity", 40.0)
|
|
|
|
self.assertEqual(["s1", "s3"], long_frame["sample_id"].tolist())
|
|
self.assertEqual(["s2", "s4"], short_frame["sample_id"].tolist())
|
|
|
|
def test_nonlinear_pm_probe_exit_metrics_describe_trade_outcomes(self) -> None:
|
|
trades = pd.DataFrame(
|
|
{
|
|
"target_hit": [1, 0, 0],
|
|
"stop_hit": [0, 1, 0],
|
|
"time_to_exit_ms": [300_000, 600_000, 2_700_000],
|
|
}
|
|
)
|
|
|
|
metrics = _exit_metrics(trades)
|
|
|
|
self.assertAlmostEqual(1 / 3, metrics["target_hit_rate"])
|
|
self.assertAlmostEqual(1 / 3, metrics["stop_hit_rate"])
|
|
self.assertAlmostEqual(1 / 3, metrics["timeout_exit_rate"])
|
|
self.assertAlmostEqual(20.0, metrics["avg_time_to_exit_min"])
|
|
self.assertAlmostEqual(10.0, metrics["p50_time_to_exit_min"])
|
|
|
|
def test_entry_feature_screen_prefers_actual_plan_edge(self) -> None:
|
|
dataset = pd.DataFrame(
|
|
{
|
|
"long_expected_net_edge_bps": [20.0],
|
|
"short_expected_net_edge_bps": [15.0],
|
|
"long_actual_plan_net_edge_bps": [-3.0],
|
|
"short_actual_plan_net_edge_bps": [4.0],
|
|
}
|
|
)
|
|
|
|
self.assertEqual("long_actual_plan_net_edge_bps", _screen_edge_column(dataset, "LONG"))
|
|
self.assertEqual("short_actual_plan_net_edge_bps", _screen_edge_column(dataset, "SHORT"))
|
|
|
|
def test_entry_feature_screen_requires_actual_plan_edge(self) -> None:
|
|
dataset = pd.DataFrame({"long_expected_net_edge_bps": [20.0]})
|
|
|
|
with self.assertRaises(ValueError):
|
|
_screen_edge_column(dataset, "LONG")
|
|
|
|
def test_entry_regression_heads_train_on_actual_plan_edge(self) -> None:
|
|
heads = {head[0]: head[2] for head in TARGETS["ENTRY"]["heads"]}
|
|
|
|
self.assertEqual("long_actual_plan_net_edge_bps", heads["long_expected_net_edge_bps"])
|
|
self.assertEqual("short_actual_plan_net_edge_bps", heads["short_expected_net_edge_bps"])
|
|
|
|
def test_conditional_entry_training_uses_direction_label_rows(self) -> None:
|
|
train = pd.DataFrame({"long_target": [1, 0, 1, 0], "short_target": [0, 1, 0, 1]})
|
|
|
|
long_mask, long_filter = _head_train_mask("ENTRY", "long_entry_prob", train, Namespace(conditional_entry_direction_labels=True))
|
|
short_mask, short_filter = _head_train_mask("ENTRY", "short_expected_net_edge_bps", train, Namespace(conditional_entry_direction_labels=True))
|
|
default_mask, default_filter = _head_train_mask("ENTRY", "long_entry_prob", train, Namespace(conditional_entry_direction_labels=False))
|
|
|
|
self.assertEqual("DIRECTION_LABEL_LONG_FIT_ROWS", long_filter)
|
|
self.assertEqual([True, False, True, False], long_mask.tolist())
|
|
self.assertEqual("DIRECTION_LABEL_SHORT_FIT_ROWS", short_filter)
|
|
self.assertEqual([False, True, False, True], short_mask.tolist())
|
|
self.assertEqual("ALL_FIT_ROWS", default_filter)
|
|
self.assertEqual([True, True, True, True], default_mask.tolist())
|
|
|
|
def test_conditional_entry_training_can_use_side_opportunity_rows(self) -> None:
|
|
train = pd.DataFrame(
|
|
{
|
|
"long_max_achievable_net_edge_bps": [45.0, 10.0, 60.0, 39.0],
|
|
"short_max_achievable_net_edge_bps": [8.0, 42.0, 15.0, 80.0],
|
|
}
|
|
)
|
|
args = Namespace(
|
|
conditional_entry_source="side_opportunity",
|
|
conditional_entry_direction_labels=False,
|
|
conditional_entry_opportunity_bps=40.0,
|
|
)
|
|
|
|
long_mask, long_filter = _head_train_mask("ENTRY", "long_entry_prob", train, args)
|
|
short_mask, short_filter = _head_train_mask("ENTRY", "short_expected_net_edge_bps", train, args)
|
|
|
|
self.assertEqual("SIDE_OPPORTUNITY_LONG_GE_40_BPS_FIT_ROWS", long_filter)
|
|
self.assertEqual([True, False, True, False], long_mask.tolist())
|
|
self.assertEqual("SIDE_OPPORTUNITY_SHORT_GE_40_BPS_FIT_ROWS", short_filter)
|
|
self.assertEqual([False, True, False, True], short_mask.tolist())
|
|
|
|
def test_direction_opportunity_labels_choose_clear_path_opportunity(self) -> None:
|
|
labels = _opportunity_labels(
|
|
np.array([45.0, 10.0, 45.0, 42.0, np.nan]),
|
|
np.array([20.0, 50.0, 43.0, 48.0, 50.0]),
|
|
opportunity_bps=40.0,
|
|
min_advantage_bps=5.0,
|
|
)
|
|
|
|
self.assertEqual([1, 0, 0, 0, 0], labels["long_target"].tolist())
|
|
self.assertEqual([0, 1, 0, 1, 1], labels["short_target"].tolist())
|
|
self.assertEqual([0, 0, 1, 0, 0], labels["neutral_target"].tolist())
|
|
|
|
def test_diagnostics_reads_actual_training_dataset_labels(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
root = Path(tmp)
|
|
dataset_dir = root / "dataset"
|
|
dataset_dir.mkdir(parents=True)
|
|
pd.DataFrame(
|
|
{
|
|
"sample_id": ["s1", "s2"],
|
|
"split_id": ["fit_inner", "fit_inner"],
|
|
"long_target": [1, 0],
|
|
"short_target": [0, 0],
|
|
"neutral_target": [0, 1],
|
|
"future_return_bps": [5.0, -1.0],
|
|
}
|
|
).to_parquet(dataset_dir / "direction_train.parquet", index=False)
|
|
pd.DataFrame(
|
|
{
|
|
"sample_id": ["s1", "s2"],
|
|
"split_id": ["fit_inner", "fit_inner"],
|
|
"long_entry_target": [1, 0],
|
|
"short_entry_target": [0, 1],
|
|
"long_actual_plan_net_edge_bps": [8.0, -6.0],
|
|
"short_actual_plan_net_edge_bps": [-5.0, 7.0],
|
|
}
|
|
).to_parquet(dataset_dir / "entry_train.parquet", index=False)
|
|
|
|
summary = _label_summary(root)
|
|
|
|
self.assertEqual("dataset/direction_train.parquet", summary["fit_inner"]["direction"]["source"])
|
|
self.assertEqual({"LONG": 0.5, "SHORT": 0.0, "NEUTRAL": 0.5}, summary["fit_inner"]["direction"]["label_ratio"])
|
|
self.assertEqual("dataset/entry_train.parquet", summary["fit_inner"]["entry"]["source"])
|
|
self.assertEqual(0.5, summary["fit_inner"]["entry"]["target_rate_by_side"]["LONG"])
|
|
|
|
def test_good_trade_structure_builds_side_frame_and_top_metrics(self) -> None:
|
|
dataset = pd.DataFrame(
|
|
{
|
|
"sample_id": ["s1", "s2", "s3"],
|
|
"split_id": ["fit_inner", "fit_inner", "fit_inner"],
|
|
"long_actual_plan_net_edge_bps": [4.0, -5.0, 1.0],
|
|
"short_actual_plan_net_edge_bps": [-5.0, 6.0, -1.0],
|
|
**{feature: [0.1, 0.2, 0.3] for feature in FEATURE_ORDER},
|
|
}
|
|
)
|
|
|
|
frame = _side_frame(dataset, "LONG", min_good_edge_bps=3.0, bad_edge_bps=-3.0)
|
|
metrics = _top_fraction_metrics(frame, np.array([0.9, 0.1, 0.2]), 1 / 3)
|
|
|
|
self.assertEqual([1, 0, 0], frame["good_trade"].tolist())
|
|
self.assertEqual([0, 1, 0], frame["bad_trade"].tolist())
|
|
self.assertEqual(1, metrics["rows"])
|
|
self.assertEqual(1.0, metrics["good_rate"])
|
|
self.assertEqual(4.0, metrics["avg_edge_bps"])
|
|
|
|
def test_entry_feature_screen_keeps_zero_inflated_event_features(self) -> None:
|
|
values = np.concatenate((np.zeros(5000), np.linspace(1.0, 100.0, 500)))
|
|
edges = _bucket_edges(values)
|
|
|
|
self.assertGreaterEqual(len(edges), 3)
|
|
self.assertEqual(-np.inf, edges[0])
|
|
self.assertEqual(np.inf, edges[-1])
|
|
|
|
def test_entry_condition_pair_screen_finds_stable_two_feature_filter(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
data_root = Path(tmp)
|
|
run_root = data_root / "trader-v4" / "runs" / "unit-condition-pair"
|
|
dataset_path = run_root / "dataset" / "entry_train.parquet"
|
|
dataset_path.parent.mkdir(parents=True)
|
|
|
|
frames = []
|
|
row_count = 1200
|
|
base_feature_values = np.linspace(0.0, 0.999, row_count)
|
|
for split_id in TRAINING_SPLITS:
|
|
frame = pd.DataFrame({feature: 0.0 for feature in FEATURE_ORDER}, index=np.arange(row_count))
|
|
frame["split_id"] = split_id
|
|
frame["ret_1m_bps"] = base_feature_values
|
|
frame["ret_5m_bps"] = base_feature_values
|
|
good_mask = (frame["ret_1m_bps"] > 0.9) & (frame["ret_5m_bps"] > 0.9)
|
|
frame["long_entry_target"] = good_mask.astype(int)
|
|
frame["short_entry_target"] = 0
|
|
frame["long_actual_plan_net_edge_bps"] = np.where(good_mask, 8.0, -6.0)
|
|
frame["short_actual_plan_net_edge_bps"] = -6.0
|
|
frame["long_mae_bps"] = np.where(good_mask, 2.0, 15.0)
|
|
frame["short_mae_bps"] = 15.0
|
|
frames.append(frame)
|
|
pd.concat(frames, ignore_index=True).to_parquet(dataset_path, index=False)
|
|
|
|
screen_entry_condition_pairs(
|
|
Namespace(
|
|
data_root=data_root,
|
|
run_id="unit-condition-pair",
|
|
min_seed_rows=50,
|
|
min_pair_rows=50,
|
|
max_seed_conditions_per_side=8,
|
|
max_buckets_per_feature=2,
|
|
)
|
|
)
|
|
|
|
result = read_json(run_root / "diagnostics" / "entry_condition_pair_screen_result.json")
|
|
candidates = pd.read_csv(run_root / "diagnostics" / "entry_condition_pair_candidates.csv")
|
|
|
|
self.assertGreater(result["stable_candidate_count"], 0)
|
|
self.assertTrue(candidates["usable_candidate"].any())
|
|
best = candidates.iloc[0]
|
|
self.assertEqual("LONG", best["side"])
|
|
self.assertGreater(float(best["min_eval_edge_bps"]), 0.0)
|
|
|
|
def test_entry_mae_label_diagnostic_finds_low_drawdown_target(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
data_root = Path(tmp)
|
|
run_root = data_root / "trader-v4" / "runs" / "unit-mae-diagnostic"
|
|
dataset_path = run_root / "dataset" / "entry_train.parquet"
|
|
dataset_path.parent.mkdir(parents=True)
|
|
|
|
frames = []
|
|
row_count = 800
|
|
base_feature_values = np.linspace(0.0, 0.999, row_count)
|
|
for split_id in TRAINING_SPLITS:
|
|
frame = pd.DataFrame({feature: 0.0 for feature in FEATURE_ORDER}, index=np.arange(row_count))
|
|
frame["split_id"] = split_id
|
|
frame["ret_1m_bps"] = base_feature_values
|
|
good_mask = frame["ret_1m_bps"] > 0.85
|
|
frame["long_entry_target"] = good_mask.astype(int)
|
|
frame["short_entry_target"] = 0
|
|
frame["long_actual_plan_net_edge_bps"] = np.where(good_mask, 9.0, -6.0)
|
|
frame["short_actual_plan_net_edge_bps"] = -6.0
|
|
frame["long_max_achievable_net_edge_bps"] = np.where(good_mask, 18.0, 2.0)
|
|
frame["short_max_achievable_net_edge_bps"] = 2.0
|
|
frame["long_mae_bps"] = np.where(good_mask, 2.0, 15.0)
|
|
frame["short_mae_bps"] = 15.0
|
|
frames.append(frame)
|
|
pd.concat(frames, ignore_index=True).to_parquet(dataset_path, index=False)
|
|
|
|
diagnose_entry_mae_labels(
|
|
Namespace(
|
|
data_root=data_root,
|
|
run_id="unit-mae-diagnostic",
|
|
max_mae_bps=(4.0,),
|
|
min_opportunity_bps=(12.0,),
|
|
model_families=("linear",),
|
|
top_fraction=0.10,
|
|
max_train_rows=0,
|
|
)
|
|
)
|
|
|
|
result = read_json(run_root / "diagnostics" / "entry_mae_label_diagnostic_result.json")
|
|
candidates = pd.read_csv(run_root / "diagnostics" / "entry_mae_label_diagnostic_candidates.csv")
|
|
|
|
self.assertGreater(result["positive_top_edge_candidate_count"], 0)
|
|
best = candidates.iloc[0]
|
|
self.assertEqual("LONG", best["side"])
|
|
self.assertTrue(bool(best["stable_top_edge_positive"]))
|
|
|
|
def test_conditional_entry_probe_finds_positive_oracle_direction_subset(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
data_root = Path(tmp)
|
|
run_root = data_root / "trader-v4" / "runs" / "unit-conditional-entry"
|
|
dataset_path = run_root / "dataset" / "entry_train.parquet"
|
|
dataset_path.parent.mkdir(parents=True)
|
|
|
|
frames = []
|
|
row_count = 900
|
|
base_feature_values = np.linspace(0.0, 0.999, row_count)
|
|
for split_id in TRAINING_SPLITS:
|
|
frame = pd.DataFrame({feature: 0.0 for feature in FEATURE_ORDER}, index=np.arange(row_count))
|
|
frame["split_id"] = split_id
|
|
frame["ret_1m_bps"] = base_feature_values
|
|
good_mask = frame["ret_1m_bps"] > 0.85
|
|
opportunity_mask = frame["ret_1m_bps"] > 0.50
|
|
frame["long_actual_plan_net_edge_bps"] = np.where(good_mask, 10.0, -6.0)
|
|
frame["short_actual_plan_net_edge_bps"] = -6.0
|
|
frame["long_max_achievable_net_edge_bps"] = np.where(opportunity_mask, 40.0, 2.0)
|
|
frame["short_max_achievable_net_edge_bps"] = 2.0
|
|
frames.append(frame)
|
|
pd.concat(frames, ignore_index=True).to_parquet(dataset_path, index=False)
|
|
|
|
probe_conditional_entry_training(
|
|
Namespace(
|
|
data_root=data_root,
|
|
run_id="unit-conditional-entry",
|
|
condition_opportunity_bps=(20.0,),
|
|
target_edge_bps=(0.0,),
|
|
model_families=("linear",),
|
|
top_fractions=(0.10,),
|
|
max_train_rows=0,
|
|
min_train_rows=50,
|
|
min_eval_rows=50,
|
|
)
|
|
)
|
|
|
|
result = read_json(run_root / "diagnostics" / "conditional_entry_probe_result.json")
|
|
candidates = pd.read_csv(run_root / "diagnostics" / "conditional_entry_probe_candidates.csv")
|
|
|
|
self.assertGreater(result["stable_positive_count"], 0)
|
|
self.assertTrue(candidates.iloc[0]["stable_positive"])
|
|
self.assertGreater(float(candidates.iloc[0]["min_top_edge_bps"]), 0.0)
|
|
|
|
def test_dynamic_exit_search_writes_plan_diagnostics(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
data_root = Path(tmp)
|
|
run_root = data_root / "trader-v4" / "runs" / "unit-dynamic-exit"
|
|
feature_path = run_root / "feature" / "feature_frame.parquet"
|
|
replay_path = run_root / "replay" / "replay_1m.parquet"
|
|
config_path = data_root / "label_config.json"
|
|
feature_path.parent.mkdir(parents=True)
|
|
replay_path.parent.mkdir(parents=True)
|
|
|
|
times = pd.date_range("2026-01-01", periods=7, freq="min", tz="UTC")
|
|
pd.DataFrame(
|
|
{
|
|
"sample_id": [f"s{i}" for i in range(4)],
|
|
"symbol": ["BTC-USDT-PERP"] * 4,
|
|
"event_time": times[:4],
|
|
"open_time_ms": [0, 60_000, 120_000, 180_000],
|
|
"split_id": ["tune_inner", "validation_locked", "latest_stress", "fit_inner"],
|
|
"walk_forward_fold": [0, 0, 0, 0],
|
|
"data_quality_flag": ["OK", "OK", "OK", "OK"],
|
|
}
|
|
).to_parquet(feature_path, index=False)
|
|
pd.DataFrame(
|
|
{
|
|
"event_time": times,
|
|
"open_time_ms": [0, 60_000, 120_000, 180_000, 240_000, 300_000, 360_000],
|
|
"symbol": ["BTC-USDT-PERP"] * 7,
|
|
"open": [100.0] * 7,
|
|
"high": [100.0, 100.12, 100.22, 100.24, 100.24, 100.24, 100.24],
|
|
"low": [100.0, 100.00, 100.00, 100.00, 100.00, 100.00, 100.00],
|
|
"close": [100.0, 100.10, 100.18, 100.20, 100.20, 100.20, 100.20],
|
|
"spread_bps": [1.0] * 7,
|
|
}
|
|
).to_parquet(replay_path, index=False)
|
|
write_json(config_path, {"entry": {"min_expected_net_edge_bps": 3.0}})
|
|
|
|
search_dynamic_exit_plans(
|
|
Namespace(
|
|
data_root=data_root,
|
|
run_id="unit-dynamic-exit",
|
|
feature_path=feature_path,
|
|
replay_path=replay_path,
|
|
label_config_path=config_path,
|
|
cost_config_path=None,
|
|
horizons=(3,),
|
|
targets=(10.0,),
|
|
stops=(5.0,),
|
|
trailing_stops=(4.0,),
|
|
second_target_multipliers=(2.0,),
|
|
take1_ratios=(0.5,),
|
|
take2_ratios=(0.25,),
|
|
output_dir_name="dynamic-exit-search",
|
|
)
|
|
)
|
|
|
|
result = read_json(run_root / "dynamic-exit-search" / "dynamic_exit_search_result.json")
|
|
self.assertEqual("DYNAMIC_TRAILING_V1", result["best_plan"]["plan_method"])
|
|
self.assertEqual(1, result["candidate_count"])
|
|
self.assertTrue((run_root / "dynamic-exit-search" / "dynamic_exit_search_report.md").is_file())
|
|
|
|
def test_ofi_entry_dataset_uses_actual_plan_edge(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
baseline_root = Path(tmp)
|
|
dataset_path = baseline_root / "dataset" / "entry_train.parquet"
|
|
dataset_path.parent.mkdir(parents=True)
|
|
pd.DataFrame(
|
|
{
|
|
"sample_id": ["s1"],
|
|
"long_entry_target": [1],
|
|
"short_entry_target": [0],
|
|
"long_actual_plan_net_edge_bps": [4.0],
|
|
"short_actual_plan_net_edge_bps": [-7.0],
|
|
}
|
|
).to_parquet(dataset_path, index=False)
|
|
feature = pd.DataFrame(
|
|
{
|
|
"sample_id": ["s1"],
|
|
"symbol": ["BTC-USDT-PERP"],
|
|
"event_time": pd.to_datetime(["2026-01-01T00:00:00Z"]),
|
|
"open_time_ms": [0],
|
|
"split_id": ["fit_inner"],
|
|
"walk_forward_fold": [0],
|
|
"data_quality_flag": ["OK"],
|
|
}
|
|
)
|
|
|
|
dataset = _load_entry_dataset(baseline_root, feature)
|
|
|
|
self.assertIn("long_actual_plan_net_edge_bps", dataset.columns)
|
|
self.assertNotIn("long_expected_net_edge_bps", dataset.columns)
|
|
self.assertEqual(4.0, float(dataset.loc[0, "long_actual_plan_net_edge_bps"]))
|
|
|
|
def test_split_builder_uses_locked_validation_contract(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
data_root = Path(tmp)
|
|
replay_path = data_root / "replay_1m.parquet"
|
|
frame = pd.DataFrame(
|
|
{
|
|
"event_time": pd.date_range("2025-06-20", "2026-06-19", freq="D", tz="UTC"),
|
|
"symbol": "BTC-USDT-PERP",
|
|
}
|
|
)
|
|
frame.to_parquet(replay_path, index=False)
|
|
|
|
build_splits(
|
|
Namespace(
|
|
data_root=data_root,
|
|
run_id="unit-split",
|
|
replay_path=replay_path,
|
|
fit_inner_start="2025-06-20",
|
|
fit_inner_end="2026-01-15",
|
|
tune_inner_start="2026-01-16",
|
|
tune_inner_end="2026-02-28",
|
|
validation_locked_start="2026-03-01",
|
|
validation_locked_end="2026-04-30",
|
|
latest_stress_start="2026-05-01",
|
|
latest_stress_end="2026-06-19",
|
|
gap_minutes=0,
|
|
fold_count=2,
|
|
)
|
|
)
|
|
|
|
manifest = read_json(data_root / "trader-v4" / "runs" / "unit-split" / "split" / "split_manifest.json")
|
|
self.assertEqual(set(TRAINING_SPLITS), {item["split_id"] for item in manifest["splits"]})
|
|
self.assertEqual([VALIDATION_LOCKED_SPLIT, LATEST_STRESS_SPLIT], manifest["sealed_splits"])
|
|
self.assertEqual("FINAL_GATE_ONLY", manifest["latest_stress_policy"])
|
|
|
|
def test_path_stats_keeps_same_bar_target_stop_as_stop_first(self) -> None:
|
|
frame = pd.DataFrame(
|
|
{
|
|
"event_time": pd.date_range("2026-01-01", periods=6, freq="min", tz="UTC"),
|
|
"open_time_ms": np.arange(6, dtype=np.int64) * 60_000,
|
|
"symbol": "BTC-USDT-PERP",
|
|
"close": [100.0, 100.0, 100.0, 100.0, 100.0, 100.0],
|
|
"high": [100.0, 100.05, 100.20, 100.0, 100.0, 100.0],
|
|
"low": [100.0, 99.95, 99.70, 100.0, 100.0, 100.0],
|
|
"spread_bps": [1.0, 1.1, 1.2, 1.3, 1.4, 1.5],
|
|
}
|
|
)
|
|
|
|
stats = _path_stats_for_group(frame, "LONG", horizon=3, target_bps=10.0, stop_bps=8.0)
|
|
first = stats.loc[stats["open_time_ms"].eq(0)].iloc[0]
|
|
|
|
self.assertEqual(0, first["target_hit"])
|
|
self.assertEqual(1, first["stop_hit"])
|
|
self.assertEqual(1, first["ambiguous_hit"])
|
|
self.assertEqual(120_000, first["time_to_stop_ms"])
|
|
self.assertAlmostEqual(-8.0, first["gross_edge_bps"])
|
|
|
|
def test_entry_label_uses_price_plan_outcome_not_max_future_edge(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
data_root = Path(tmp)
|
|
run_root = data_root / "trader-v4" / "runs" / "unit-entry"
|
|
feature_path = run_root / "feature" / "feature_frame.parquet"
|
|
replay_path = run_root / "replay" / "replay_1m.parquet"
|
|
plan_path = run_root / "label" / "price_plan_context.json"
|
|
config_path = data_root / "label_config.json"
|
|
feature_path.parent.mkdir(parents=True)
|
|
replay_path.parent.mkdir(parents=True)
|
|
|
|
times = pd.date_range("2026-01-01", periods=5, freq="min", tz="UTC")
|
|
pd.DataFrame(
|
|
{
|
|
"sample_id": ["s0", "s1"],
|
|
"symbol": "BTC-USDT-PERP",
|
|
"event_time": times[:2],
|
|
"open_time_ms": [0, 60_000],
|
|
"split_id": "fit_inner",
|
|
"walk_forward_fold": 0,
|
|
"data_quality_flag": "OK",
|
|
"spread_bps": 1.0,
|
|
"spread_rank_24h_pct": 0.1,
|
|
"realized_vol_15m_bps": 2.0,
|
|
}
|
|
).to_parquet(feature_path, index=False)
|
|
pd.DataFrame(
|
|
{
|
|
"event_time": times,
|
|
"open_time_ms": np.arange(5, dtype=np.int64) * 60_000,
|
|
"symbol": "BTC-USDT-PERP",
|
|
"open": [100.0, 100.0, 100.0, 100.0, 100.0],
|
|
"high": [100.0, 100.05, 100.19, 100.20, 100.0],
|
|
"low": [100.0, 99.99, 99.98, 99.97, 100.0],
|
|
"close": [100.0, 100.0, 100.0, 100.0, 100.0],
|
|
"spread_bps": 1.0,
|
|
}
|
|
).to_parquet(replay_path, index=False)
|
|
write_json(
|
|
config_path,
|
|
{
|
|
"entry": {
|
|
"max_hold_minutes": 3,
|
|
"target_bps": 50.0,
|
|
"stop_bps": 50.0,
|
|
"min_expected_net_edge_bps": 3.0,
|
|
}
|
|
},
|
|
)
|
|
write_json(
|
|
plan_path,
|
|
{
|
|
"pricePlanId": "unit-plan",
|
|
"pricePlanConfigHash": "unit-hash",
|
|
"targetDistanceBps": 50.0,
|
|
"stopDistanceBps": 50.0,
|
|
"maxHoldMinutes": 3,
|
|
"costBps": 6.5,
|
|
"entryLabelMethod": ENTRY_LABEL_METHOD,
|
|
},
|
|
)
|
|
|
|
build_entry_labels(
|
|
Namespace(
|
|
data_root=data_root,
|
|
run_id="unit-entry",
|
|
feature_path=feature_path,
|
|
replay_path=replay_path,
|
|
label_config_path=config_path,
|
|
cost_config_path=None,
|
|
price_plan_context_path=plan_path,
|
|
)
|
|
)
|
|
|
|
labels = pd.read_parquet(run_root / "label" / "entry_labels.parquet")
|
|
row = labels[labels["sample_id"].eq("s0") & labels["side"].eq("LONG")].iloc[0]
|
|
self.assertEqual(0, row["target_hit"])
|
|
self.assertEqual(0, row["entry_target"])
|
|
self.assertEqual(ENTRY_LABEL_METHOD, row["label_method"])
|
|
self.assertAlmostEqual(-6.5, row["expected_net_edge_bps"], places=6)
|
|
self.assertAlmostEqual(row["gross_edge_bps"] - row["cost_bps"], row["expected_net_edge_bps"], places=6)
|
|
self.assertAlmostEqual(row["mfe_bps"] - row["cost_bps"], row["max_achievable_net_edge_bps"], places=6)
|
|
|
|
def test_entry_opportunity_label_keeps_plan_outcome_for_pm(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
data_root = Path(tmp)
|
|
run_root = data_root / "trader-v4" / "runs" / "unit-entry-opportunity"
|
|
feature_path = run_root / "feature" / "feature_frame.parquet"
|
|
replay_path = run_root / "replay" / "replay_1m.parquet"
|
|
plan_path = run_root / "label" / "price_plan_context.json"
|
|
config_path = data_root / "label_config.json"
|
|
feature_path.parent.mkdir(parents=True)
|
|
replay_path.parent.mkdir(parents=True)
|
|
|
|
times = pd.date_range("2026-01-01", periods=5, freq="min", tz="UTC")
|
|
pd.DataFrame(
|
|
{
|
|
"sample_id": ["s0"],
|
|
"symbol": "BTC-USDT-PERP",
|
|
"event_time": [times[0]],
|
|
"open_time_ms": [0],
|
|
"split_id": "fit_inner",
|
|
"walk_forward_fold": 0,
|
|
"data_quality_flag": "OK",
|
|
"spread_bps": 1.0,
|
|
"spread_rank_24h_pct": 0.1,
|
|
"realized_vol_15m_bps": 2.0,
|
|
}
|
|
).to_parquet(feature_path, index=False)
|
|
pd.DataFrame(
|
|
{
|
|
"event_time": times,
|
|
"open_time_ms": np.arange(5, dtype=np.int64) * 60_000,
|
|
"symbol": "BTC-USDT-PERP",
|
|
"open": [100.0] * 5,
|
|
"high": [100.0, 100.05, 100.19, 100.20, 100.0],
|
|
"low": [100.0, 99.99, 99.98, 99.97, 100.0],
|
|
"close": [100.0] * 5,
|
|
"spread_bps": 1.0,
|
|
}
|
|
).to_parquet(replay_path, index=False)
|
|
write_json(
|
|
config_path,
|
|
{
|
|
"entry": {
|
|
"max_hold_minutes": 3,
|
|
"target_bps": 50.0,
|
|
"stop_bps": 50.0,
|
|
"min_expected_net_edge_bps": 3.0,
|
|
"target_method": "OPPORTUNITY_MFE_V1",
|
|
}
|
|
},
|
|
)
|
|
write_json(
|
|
plan_path,
|
|
{
|
|
"pricePlanId": "unit-plan",
|
|
"pricePlanConfigHash": "unit-hash",
|
|
"targetDistanceBps": 50.0,
|
|
"stopDistanceBps": 50.0,
|
|
"maxHoldMinutes": 3,
|
|
"costBps": 6.5,
|
|
"entryLabelMethod": ENTRY_LABEL_METHOD,
|
|
"entryTargetMethod": "OPPORTUNITY_MFE_V1",
|
|
},
|
|
)
|
|
|
|
build_entry_labels(
|
|
Namespace(
|
|
data_root=data_root,
|
|
run_id="unit-entry-opportunity",
|
|
feature_path=feature_path,
|
|
replay_path=replay_path,
|
|
label_config_path=config_path,
|
|
cost_config_path=None,
|
|
price_plan_context_path=plan_path,
|
|
)
|
|
)
|
|
|
|
labels = pd.read_parquet(run_root / "label" / "entry_labels.parquet")
|
|
row = labels[labels["sample_id"].eq("s0") & labels["side"].eq("LONG")].iloc[0]
|
|
self.assertEqual(0, row["target_hit"])
|
|
self.assertEqual(1, row["entry_target"])
|
|
self.assertEqual("OPPORTUNITY_MFE_V1", row["label_method"])
|
|
self.assertAlmostEqual(row["mfe_bps"] - row["cost_bps"], row["expected_net_edge_bps"], places=6)
|
|
self.assertAlmostEqual(-6.5, row["gross_edge_bps"] - row["cost_bps"], places=6)
|
|
|
|
def test_entry_quality_label_rejects_untradable_opportunity(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
data_root = Path(tmp)
|
|
run_root = data_root / "trader-v4" / "runs" / "unit-entry-quality"
|
|
feature_path = run_root / "feature" / "feature_frame.parquet"
|
|
replay_path = run_root / "replay" / "replay_1m.parquet"
|
|
plan_path = run_root / "label" / "price_plan_context.json"
|
|
config_path = data_root / "label_config.json"
|
|
feature_path.parent.mkdir(parents=True)
|
|
replay_path.parent.mkdir(parents=True)
|
|
|
|
times = pd.date_range("2026-01-01", periods=5, freq="min", tz="UTC")
|
|
pd.DataFrame(
|
|
{
|
|
"sample_id": ["s0"],
|
|
"symbol": "BTC-USDT-PERP",
|
|
"event_time": [times[0]],
|
|
"open_time_ms": [0],
|
|
"split_id": "fit_inner",
|
|
"walk_forward_fold": 0,
|
|
"data_quality_flag": "OK",
|
|
"spread_bps": 1.0,
|
|
"spread_rank_24h_pct": 0.1,
|
|
"realized_vol_15m_bps": 2.0,
|
|
}
|
|
).to_parquet(feature_path, index=False)
|
|
pd.DataFrame(
|
|
{
|
|
"event_time": times,
|
|
"open_time_ms": np.arange(5, dtype=np.int64) * 60_000,
|
|
"symbol": "BTC-USDT-PERP",
|
|
"open": [100.0] * 5,
|
|
"high": [100.0, 100.05, 100.19, 100.20, 100.0],
|
|
"low": [100.0, 99.99, 99.98, 99.97, 100.0],
|
|
"close": [100.0] * 5,
|
|
"spread_bps": 1.0,
|
|
}
|
|
).to_parquet(replay_path, index=False)
|
|
write_json(
|
|
config_path,
|
|
{
|
|
"entry": {
|
|
"max_hold_minutes": 3,
|
|
"target_bps": 50.0,
|
|
"stop_bps": 50.0,
|
|
"min_expected_net_edge_bps": 3.0,
|
|
"min_plan_net_edge_bps": 0.0,
|
|
"max_entry_mae_bps": 12.0,
|
|
"target_method": "OPPORTUNITY_QUALITY_V1",
|
|
}
|
|
},
|
|
)
|
|
write_json(
|
|
plan_path,
|
|
{
|
|
"pricePlanId": "unit-plan",
|
|
"pricePlanConfigHash": "unit-hash",
|
|
"targetDistanceBps": 50.0,
|
|
"stopDistanceBps": 50.0,
|
|
"maxHoldMinutes": 3,
|
|
"costBps": 6.5,
|
|
"entryLabelMethod": ENTRY_LABEL_METHOD,
|
|
"entryTargetMethod": "OPPORTUNITY_QUALITY_V1",
|
|
},
|
|
)
|
|
|
|
build_entry_labels(
|
|
Namespace(
|
|
data_root=data_root,
|
|
run_id="unit-entry-quality",
|
|
feature_path=feature_path,
|
|
replay_path=replay_path,
|
|
label_config_path=config_path,
|
|
cost_config_path=None,
|
|
price_plan_context_path=plan_path,
|
|
)
|
|
)
|
|
|
|
labels = pd.read_parquet(run_root / "label" / "entry_labels.parquet")
|
|
row = labels[labels["sample_id"].eq("s0") & labels["side"].eq("LONG")].iloc[0]
|
|
self.assertEqual("OPPORTUNITY_QUALITY_V1", row["label_method"])
|
|
self.assertGreater(row["expected_net_edge_bps"], 3.0)
|
|
self.assertLess(row["actual_plan_net_edge_bps"], 0.0)
|
|
self.assertEqual(0, row["entry_target"])
|
|
|
|
def test_l1_snapshot_diff_ofi_uses_quote_notional_and_signed_ask_side(self) -> None:
|
|
bid_part, ask_part = l1_snapshot_diff_ofi_quote(
|
|
pd.Series([101.0, 101.0, 100.5]),
|
|
pd.Series([2.0, 3.0, 4.0]),
|
|
pd.Series([102.0, 101.5, 102.5]),
|
|
pd.Series([5.0, 6.0, 7.0]),
|
|
pd.Series([100.0, 101.0, 101.0]),
|
|
pd.Series([1.5, 2.0, 3.0]),
|
|
pd.Series([102.0, 102.0, 101.5]),
|
|
pd.Series([4.0, 5.0, 6.0]),
|
|
)
|
|
|
|
self.assertAlmostEqual(202.0, bid_part.iloc[0])
|
|
self.assertAlmostEqual(-102.0, ask_part.iloc[0])
|
|
self.assertAlmostEqual(101.0, bid_part.iloc[1])
|
|
self.assertAlmostEqual(-609.0, ask_part.iloc[1])
|
|
self.assertAlmostEqual(-303.0, bid_part.iloc[2])
|
|
self.assertAlmostEqual(609.0, ask_part.iloc[2])
|
|
|
|
def test_exported_onnx_accepts_java_feature_shape(self) -> None:
|
|
import onnxruntime as ort
|
|
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
path = Path(tmp) / "direction.onnx"
|
|
export_heads(
|
|
path,
|
|
[
|
|
LinearHead(
|
|
"direction",
|
|
"softmax",
|
|
np.zeros((len(FEATURE_ORDER), 3), dtype=np.float32),
|
|
np.array([0.1, 0.2, 0.3], dtype=np.float32),
|
|
)
|
|
],
|
|
feature_count=len(FEATURE_ORDER),
|
|
)
|
|
session = ort.InferenceSession(str(path))
|
|
output = session.run(None, {"features": np.zeros((1, len(FEATURE_ORDER)), dtype=np.float32)})[0]
|
|
self.assertEqual((1, 3), output.shape)
|
|
self.assertAlmostEqual(1.0, float(output.sum()), places=6)
|
|
|
|
def test_promotion_requires_passed_validation_and_marks_all_manifests_active(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
root = Path(tmp) / "artifact_bundle"
|
|
manifest_dir = root / "manifests"
|
|
manifest_dir.mkdir(parents=True)
|
|
write_json(root.parent / "artifact_validation_result.json", {"status": "PASS", "release_gate_status": "PASS", "release_gate_reasons": []})
|
|
write_json(manifest_dir / "model_bundle_manifest.json", {"status": "CANDIDATE"})
|
|
write_json(manifest_dir / "model_manifest.json", [{"model_name": "DIRECTION", "status": "CANDIDATE"}])
|
|
write_json(manifest_dir / "calibration_manifest.json", [{"model_name": "DIRECTION", "status": "CANDIDATE"}])
|
|
write_json(manifest_dir / "position_manager_manifest.json", {"status": "CANDIDATE"})
|
|
write_json(manifest_dir / "training_export_manifest.json", {"status": "CANDIDATE"})
|
|
|
|
promote_artifact_bundle(Namespace(artifact_root=root, reason="unit test"))
|
|
|
|
self.assertEqual("ACTIVE", read_json(manifest_dir / "model_bundle_manifest.json")["status"])
|
|
self.assertEqual("ACTIVE", read_json(manifest_dir / "model_manifest.json")[0]["status"])
|
|
self.assertEqual("ACTIVE", read_json(manifest_dir / "calibration_manifest.json")[0]["status"])
|
|
self.assertEqual("ACTIVE", read_json(manifest_dir / "position_manager_manifest.json")["status"])
|
|
self.assertEqual("ACTIVE", read_json(manifest_dir / "training_export_manifest.json")["status"])
|
|
|
|
def test_promotion_refuses_failed_validation(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
root = Path(tmp) / "artifact_bundle"
|
|
(root / "manifests").mkdir(parents=True)
|
|
write_json(root.parent / "artifact_validation_result.json", {"status": "FAIL"})
|
|
with self.assertRaises(SystemExit):
|
|
promote_artifact_bundle(Namespace(artifact_root=root, reason="unit test"))
|
|
result = read_json(root.parent / "artifact_promotion_result.json")
|
|
self.assertEqual("REFUSED", result["status"])
|
|
self.assertEqual("validation result is not PASS", result["message"])
|
|
|
|
def test_promotion_refuses_failed_release_gate_and_overwrites_stale_result(self) -> None:
|
|
with tempfile.TemporaryDirectory() as tmp:
|
|
root = Path(tmp) / "artifact_bundle"
|
|
(root / "manifests").mkdir(parents=True)
|
|
write_json(root.parent / "artifact_promotion_result.json", {"status": "ACTIVE"})
|
|
write_json(
|
|
root.parent / "artifact_validation_result.json",
|
|
{
|
|
"status": "PASS",
|
|
"release_gate_status": "REJECTED",
|
|
"release_gate_reasons": ["backtest_status=REJECTED"],
|
|
},
|
|
)
|
|
|
|
with self.assertRaises(SystemExit):
|
|
promote_artifact_bundle(Namespace(artifact_root=root, reason="unit test"))
|
|
|
|
result = read_json(root.parent / "artifact_promotion_result.json")
|
|
self.assertEqual("REFUSED", result["status"])
|
|
self.assertEqual("release gate is not PASS", result["message"])
|
|
self.assertEqual(["backtest_status=REJECTED"], result["release_gate_reasons"])
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|