Datasets:
Tasks:
Tabular Classification
Formats:
parquet
Languages:
English
Size:
< 1K
Tags:
economics
quantitative-finance
causal-inference
macroeconomics
housing-economics
market-microstructure
License:
File size: 7,893 Bytes
ebcde1f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 | """Deterministic, presentation-only model comparison tables."""
from __future__ import annotations
import math
from collections.abc import Mapping, Sequence
from typing import Any
from microstructure.reporting.bundle import RunBundle
_JOIN_KEYS = ("instrument", "model", "horizon_events", "split", "study_date")
_TEST_SPLITS = frozenset({"test", "final_test", "holdout", "held_out"})
def _first(row: Mapping[str, Any], names: Sequence[str], default: Any = None) -> Any:
for name in names:
if name in row:
return row[name]
return default
def _canonical_row(row: Mapping[str, Any]) -> dict[str, Any]:
result = dict(row)
result["instrument"] = _first(row, ("instrument", "symbol"), "N/A")
result["model"] = _first(row, ("model", "model_name"), "N/A")
result["horizon_events"] = _first(
row, ("horizon_events", "label_horizon_events", "horizon"), "N/A"
)
# Missing split provenance is not evidence that a row is held out. Keep the
# presentation layer fail-closed so an unsplit metric cannot be promoted to
# final-test evidence merely by being serialized.
result["split"] = str(_first(row, ("split", "evaluation_split"), "unknown"))
# A multi-date held-out study can legitimately emit the same model/split
# combination more than once. Preserve the declared study date in the
# presentation join so a replication period cannot overwrite the primary
# test period. Legacy single-period rows retain one explicit sentinel and
# therefore keep their previous join behaviour.
result["study_date"] = str(_first(row, ("study_date", "period_date", "date"), "N/A"))
result["n_obs"] = _first(row, ("n_obs", "observations", "sample_size"))
result["period_start_utc"] = _first(row, ("period_start_utc", "test_start_utc", "start_utc"))
result["period_end_utc"] = _first(row, ("period_end_utc", "test_end_utc", "end_utc"))
result["brier_score"] = _first(row, ("brier_score", "brier"))
result["expected_calibration_error"] = _first(row, ("expected_calibration_error", "ece"))
result["fees_bps"] = _first(row, ("fees_bps", "cost_bps", "costs_bps"))
result["max_drawdown"] = _first(row, ("max_drawdown", "max_drawdown_units"))
return result
def _join_key(row: Mapping[str, Any]) -> tuple[str, ...]:
return tuple(str(row.get(key, "N/A")) for key in _JOIN_KEYS)
def comparison_rows(bundle: RunBundle) -> tuple[Mapping[str, Any], ...]:
"""Join predictive and execution results without recomputing any metric."""
joined: dict[tuple[str, ...], dict[str, Any]] = {}
for source_row in bundle.predictive_metrics:
row = _canonical_row(source_row)
if row["split"].lower() not in _TEST_SPLITS:
continue
joined[_join_key(row)] = row
for source_row in bundle.execution_metrics:
row = _canonical_row(source_row)
if row["split"].lower() not in _TEST_SPLITS:
continue
key = _join_key(row)
if key in joined:
joined[key].update({name: value for name, value in row.items() if value is not None})
else:
joined[key] = row
return tuple(joined[key] for key in sorted(joined))
def _number(value: Any) -> float | None:
if value is None or isinstance(value, bool):
return None
try:
number = float(value)
except (TypeError, ValueError):
return None
return number if math.isfinite(number) else None
def _format_plain(value: Any) -> str:
if value is None or value == "":
return "N/A"
return str(value).replace("|", "\\|").replace("\n", " ")
def _format_integer(value: Any) -> str:
number = _number(value)
return f"{int(number):,}" if number is not None and number.is_integer() else "N/A"
def _format_estimate(
row: Mapping[str, Any], key: str, *, decimals: int, percent: bool = False
) -> str:
estimate = _number(row.get(key))
if estimate is None:
return "N/A"
scale = 100.0 if percent else 1.0
suffix = "%" if percent else ""
rendered = f"{estimate * scale:.{decimals}f}{suffix}"
lower = _number(row.get(f"{key}_ci_low"))
upper = _number(row.get(f"{key}_ci_high"))
if lower is not None and upper is not None:
rendered += f" [{lower * scale:.{decimals}f}, {upper * scale:.{decimals}f}]{suffix}"
return rendered
def _period(row: Mapping[str, Any]) -> str:
start = _format_plain(row.get("period_start_utc"))
end = _format_plain(row.get("period_end_utc"))
if start == "N/A" and end == "N/A":
return "N/A"
return f"{start} → {end}"
def render_model_comparison(bundle: RunBundle) -> str:
"""Render a Markdown table from serialized held-out metrics only."""
git = bundle.provenance.get("git", {})
if not isinstance(git, Mapping):
git = {}
input_hashes = bundle.provenance.get("input_manifest_sha256", [])
input_hash_text = (
", ".join(str(value) for value in input_hashes)
if isinstance(input_hashes, Sequence) and not isinstance(input_hashes, str)
else _format_plain(input_hashes)
)
lines = [
f"> **{bundle.watermark}**",
"",
(
f"Run `{bundle.run_id}`; observed UTC period "
f"`{bundle.observed_start_utc}` to `{bundle.observed_end_utc}`."
),
"",
f"Configuration SHA-256: `{_format_plain(bundle.provenance.get('config_sha256'))}`.",
f"Input manifest SHA-256: `{input_hash_text or 'none'}`.",
(
f"Git commit: `{_format_plain(git.get('commit'))}`; dirty at run time: "
f"`{str(git.get('dirty')).lower()}`."
),
"",
]
rows = comparison_rows(bundle)
if not rows:
lines.extend(
[
"No held-out model-comparison rows were serialized in this run bundle.",
"",
]
)
return "\n".join(lines)
headers = (
"Instrument",
"Horizon",
"Model",
"Split",
"N",
"Test period (UTC)",
"ROC-AUC",
"PR-AUC",
"Log loss",
"Brier",
"ECE",
"Gross bps",
"Fees bps",
"Net bps",
"Fill rate",
"Turnover",
"Max drawdown",
"Selected on",
)
lines.append("| " + " | ".join(headers) + " |")
lines.append("| " + " | ".join("---" for _ in headers) + " |")
for row in rows:
values = (
_format_plain(row.get("instrument")),
_format_plain(row.get("horizon_events")),
_format_plain(row.get("model")),
_format_plain(row.get("split")),
_format_integer(row.get("n_obs")),
_period(row),
_format_estimate(row, "roc_auc", decimals=4),
_format_estimate(row, "pr_auc", decimals=4),
_format_estimate(row, "log_loss", decimals=4),
_format_estimate(row, "brier_score", decimals=4),
_format_estimate(row, "expected_calibration_error", decimals=4),
_format_estimate(row, "gross_bps", decimals=3),
_format_estimate(row, "fees_bps", decimals=3),
_format_estimate(row, "net_bps", decimals=3),
_format_estimate(row, "fill_rate", decimals=1, percent=True),
_format_estimate(row, "turnover", decimals=3),
_format_estimate(row, "max_drawdown", decimals=3),
_format_plain(row.get("selected_on")),
)
lines.append("| " + " | ".join(values) + " |")
lines.extend(
[
"",
(
"`N/A` means the producer did not serialize a comparable metric; "
"it is never interpreted as zero. Confidence intervals, when supplied, "
"are shown in brackets."
),
"",
]
)
return "\n".join(lines)
|