policy: LabelWeightSupportPolicy,
reasons: list[str],
) -> NDArray[np.floating]:
+ """Apply the configured ``support_policy`` to ``base_weights``.
+
+ ``policy='raise'`` raises ``ValueError`` with ``context`` and
+ ``reasons``. ``policy='fallback'`` logs a ``WARNING`` and returns
+ the sanitized base weights via
+ ``compose_sample_weights(base_weights, None, ...)``.
+ """
reason_text = "; ".join(reasons)
match policy:
case "raise":
*,
context: str,
) -> NDArray[np.floating]:
+ """Compose training-set weights with label-weight support gating.
+
+ When ``label_weights`` is ``None`` and
+ ``label_weighting_config['strategy'] != 'none'``, routes through
+ ``_apply_support_policy`` so the configured ``support_policy``
+ governs the outcome. Otherwise composes the label-weighted
+ product via ``compose_sample_weights`` and validates the
+ ``summarize_label_weight_support`` summary against
+ ``min_pivot_equivalent_count``,
+ ``min_positive_label_weight_fraction``, and
+ ``min_effective_sample_size``; any failure routes through
+ ``_apply_support_policy``. Returns the composed weights on
+ success or the fallback weights from the policy on failure.
+ """
policy = cast(
LabelWeightSupportPolicy, label_weighting_config["support_policy"]
)
label_weights = unfiltered_df.loc[
features_filtered.index, weight_col
].to_numpy(dtype=float)
- logger.debug(f"label weight column active: {weight_col!r}")
+ logger.debug("label weight column active: %r", weight_col)
else:
label_weights = None
logger.debug(
- f"label weight column absent ({weight_col!r}); using base weights only"
+ "label weight column absent (%r); using base weights only",
+ weight_col,
)
return SampleWeightInputs(
base=base_weights,
self._optuna_label_candles[pair] = 0
else:
logger.debug(
- f"[{pair}] Optuna {namespace} callback throttled, still {optuna_label_remaining_candles} candles to go"
+ f"[{pair}] Optuna {namespace} callback throttled, {optuna_label_remaining_candles} candles until next emission"
)
if len(self._optuna_label_incremented_pairs) >= len(self.pairs):
self._optuna_label_incremented_pairs = []
)
if candles_diff < 0:
logger.warning(
- f"[{pair}] Fit live predictions not warmed up yet, still {abs(candles_diff)} candles to go"
+ f"[{pair}] Fit live predictions not warmed up: {abs(candles_diff)} candles until warmup completion"
)
warmed_up = False
# ``np.isfinite(cutoff)`` guard substitutes
# ``_DI_CUTOFF_DEFAULT``.
fallback=(0.0, 0.0, 0.0),
- context=f"di_values_weibull_fit:{pair}",
+ context=f"[{pair}] di_values_weibull_fit",
logger=logger,
min_count=2,
require_variance=True,
sample,
sp.stats.norm.fit,
fallback=fallback,
- context=f"label_norm_fit:{pair}:{label_col}",
+ context=f"[{pair}] label_norm_fit:{label_col}",
logger=logger,
min_count=2,
require_variance=True,
EXTREMA_WEIGHT_COLUMN: Final[str] = "extrema_weight"
EXTREMA_WEIGHT_SMOOTHED_COLUMN: Final[str] = "extrema_weight_smoothed"
-LABEL_WEIGHT_SUFFIX: Final[str] = "_weight"
+_LABEL_WEIGHT_SUFFIX: Final[str] = "_weight"
_LABEL_KNOWN_AT_LOOKAHEAD_SUFFIX: Final[str] = "_known_at_lookahead"
LABEL_COLUMNS: Final[tuple[str, ...]] = (EXTREMA_COLUMN,)
nor with ``find_features`` (which selects columns containing ``%``).
Preserves the project convention where a leading ``s`` denotes a smoothed
target series (e.g. ``&s-extrema``); no ``s`` denotes a raw target.
- Raises ``ValueError`` if the result still contains ``&`` or ``%``.
+ Raises ``ValueError`` when the result contains ``&`` or ``%`` after
+ sigil strip.
Examples:
``("&s-extrema", "_weight")`` -> ``"s-extrema_weight"``
def label_weight_column_name(label_col: str) -> str:
"""Return the weight column name for a label column."""
- return _label_aux_column_name(label_col, LABEL_WEIGHT_SUFFIX)
+ return _label_aux_column_name(label_col, _LABEL_WEIGHT_SUFFIX)
def label_known_at_lookahead_column_name(label_col: str) -> str:
Detects the canonical ``(dataframe, params, logger) -> LabelData``
shape by a positional parameter named ``logger`` at index 2 (with or
- without a default). Generators without such a parameter are wrapped
- to drop the logger argument at dispatch; defaulted positionals after
- index 1 stay at their defaults. ``*args``, ``**kwargs``, keyword-only
- ``logger``, fewer than 2 required positionals, more than 3 required
- positionals, and 3 required positionals whose third name is not
- ``logger`` raise ``ValueError`` at registration. Inspection runs
- once at registration; dispatch in ``generate_label_data`` is a
- direct call.
+ without a default). Generators with exactly 2 positional parameters
+ are wrapped to drop the logger argument at dispatch. Generators with
+ a 3rd positional parameter whose name is not ``logger`` raise
+ ``ValueError`` at registration, regardless of whether that parameter
+ is required or has a default. ``*args``, ``**kwargs``, keyword-only
+ ``logger``, fewer than 2 required positionals, and more than 3
+ required positionals also raise ``ValueError``. Inspection runs once
+ at registration; dispatch in ``generate_label_data`` is a direct
+ call.
"""
sig = inspect.signature(generator)
params = list(sig.parameters.values())
f"required positional parameter(s); expected 2 "
f"(dataframe, params) or 3 (dataframe, params, logger)"
)
- has_logger_at_third = n_total >= 3 and positional[2].name == "logger"
- if has_logger_at_third:
+ if n_total >= 3:
+ if positional[2].name != "logger":
+ raise ValueError(
+ f"Invalid label generator {generator!r}: third positional "
+ f"parameter is named {positional[2].name!r}, expected "
+ f"``logger``"
+ )
return cast(LabelGenerator, generator)
- if n_required == 3:
- raise ValueError(
- f"Invalid label generator {generator!r}: third positional "
- f"parameter is named {positional[2].name!r}, expected "
- f"``logger``"
- )
@functools.wraps(generator)
def adapted(
if old_section == new_section:
logger.warning(f"{old_path!r} is deprecated, use {new_key!r} instead")
else:
- logger.warning(f"{old_path!r} has moved to {new_path!r}")
+ logger.warning(f"{old_path!r} is deprecated, use {new_path!r} instead")
else:
_delete_path(config, old_path)
if old_section == new_section:
Non-finite or non-positive entries are treated as ``0``; rows in
``drop_mask`` are forced to ``0``. On collapse (no positive finite
entry survives), returns ones on surviving rows and zeros on dropped
- rows, rescaled so ``mean(out) == 1`` still holds.
+ rows, rescaled so ``mean(out) == 1``.
``context`` is the caller-supplied prefix attached to every warning
and error emitted from this helper.
*,
expected_selection_metadata: dict[str, Any] | None = None,
) -> dict[str, Any] | None:
+ """Validate an Optuna ``label`` best-params payload against the v2 schema.
+
+ Returns the inner ``params`` dict on success; returns ``None`` on
+ rejection. Rejects non-dict input, missing or invalid ``schema_version``,
+ schema-version mismatch with ``_OPTUNA_LABEL_BEST_PARAMS_SCHEMA_VERSION``,
+ missing or invalid ``selection_metadata``, missing or invalid
+ ``selection_metadata.schema_version``, schema-version mismatch with
+ ``_OPTUNA_LABEL_SELECTION_SCHEMA_VERSION``, missing or invalid
+ ``label_period_candles`` / ``label_natr_multiplier`` /
+ ``label_horizon_candles``, and -- when ``expected_selection_metadata``
+ is provided -- any drift between the stored and the caller's current
+ ``selection_metadata``. Every rejection emits a ``[<pair>]``-prefixed
+ warning when ``logger`` is provided.
+ """
if not isinstance(best_params, dict):
if logger is not None:
logger.warning(f"[{pair}] Ignoring Optuna label best params: not a dict")
repositories / freqai-strategies.git / commitdiff
