multitask.predict.PredictTask

multitask.predict.PredictTask(
    config=None,
    *,
    dataframe=None,
    data_test=None,
    cache_home=None,
    log_level=logging.INFO,
    **overrides,
)

Task 5 — Predict-only using previously saved models.

Loads fitted forecasters that were persisted by a prior LazyTask or DefaultsTask run (or by any task in the spotforecast2 sibling package) and produces predictions for all configured targets. No training or tuning is performed.

If no saved models exist in the cache directory the run method raises RuntimeError with an informative message.

Examples

import tempfile
from pathlib import Path
from spotforecast2_safe.multitask import PredictTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(data_frame_name="demo10", predict_size=24, cache_home=Path(tmp))
    task = PredictTask(cfg)
    print(f"Task: {task.TASK}")
    print(f"Predict size: {task.config.predict_size}")

Task: predict
Predict size: 24

Methods

Name	Description
agg_predictor	Aggregate per-target prediction packages into a weighted forecast.
build_exogenous_features	Build, combine, encode, and merge exogenous feature covariates.
create_forecaster	Create a fresh forecaster for the given target.
cv_ts	Build a `TimeSeriesFold` for cross-validation.
detect_outliers	Apply hard-bound filtering and IsolationForest outlier detection.
impute	Fill missing values using the configured imputation strategy.
load_models	Load the most recent fitted models from the cache directory.
load_tuning_results	Load the most recent tuning results for a target from cache.
log_summary	Log a summary of the current pipeline configuration.
plot_with_outliers	Visualise original vs. cleaned data with outlier markers.
prepare_data	Load, resample, validate, and configure the pipeline data.
run	Run prediction using previously saved models.
save_models	Save fitted forecaster models to the cache directory.
save_tuning_results	Save tuning results (best parameters and lags) to a JSON file.

agg_predictor

multitask.predict.PredictTask.agg_predictor(results, targets, weights)

Aggregate per-target prediction packages into a weighted forecast.

Delegates to the module-level agg_predictor function. Available as an instance method so that subclasses can override the aggregation strategy when needed.

Parameters

Name	Type	Description	Default
results	Dict[str, Dict[str, Any]]	Mapping of target name to prediction package (as returned by `build_prediction_package`).	required
targets	List[str]	Ordered list of target names to include.	required
weights	List[float]	Per-target aggregation weights aligned with `targets`.	required

Returns

Name	Type	Description
	Dict[str, Any]	Aggregated prediction package dict.

Examples

import tempfile
import numpy as np
import pandas as pd
from spotforecast2_safe.multitask import LazyTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

rng = np.random.default_rng(0)
idx_train = pd.date_range("2023-01-01", periods=48, freq="h", tz="UTC")
idx_future = pd.date_range("2023-01-03", periods=6, freq="h", tz="UTC")

def _pkg(train_val, future_val):
    return {
        "train_actual": pd.Series(np.full(48, train_val), index=idx_train),
        "train_pred": pd.Series(np.full(48, train_val * 0.99), index=idx_train),
        "future_pred": pd.Series(np.full(6, future_val), index=idx_future),
        "future_actual": pd.Series(dtype="float64"),
    }

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(cache_home=tmp, verbose=False)
    task = LazyTask(cfg)
    results = {"wind": _pkg(100.0, 110.0), "solar": _pkg(200.0, 210.0)}
    agg = task.agg_predictor(results, ["wind", "solar"], [0.4, 0.6])
    print(f"Weighted future_pred: {agg['future_pred'].iloc[0]:.1f}")

Weighted future_pred: 170.0

build_exogenous_features

multitask.predict.PredictTask.build_exogenous_features()

Build, combine, encode, and merge exogenous feature covariates.

This is step 4-7 of the pipeline (run after prepare_data, detect_outliers, and impute). It assembles the full exogenous-covariate matrix that the forecaster consumes, then merges it onto the target data. The orchestration proceeds in order:

4a — Weather, via get_weather_features (Open-Meteo). The response is parquet-cached only when config.cache_home is set. Fetch failures are handled per config.on_weather_failure: "raise" re-raises WeatherFetchError; "skip" logs a warning and continues with an empty weather frame (fail-safe). When config.zone_weather_columns is set (opt-in, mutually exclusive with per_zone_weather / use_population_weighted_weather), this step instead calls weather.zone_columns.build_zone_weather_columns and concatenates population-weighted weather for each of the four German TSO zones as __<zone_short>-suffixed columns (e.g. temperature_2m__50hertz) for the single target.
4b — Calendar features, via get_calendar_features.
4c — Day/night (solar) features, via get_day_night_features (computed with astral from config.latitude / config.longitude).
4d — Holiday features, via get_holiday_features for config.country_code / config.state.
5 — The four frames are concatenated along the columns and any residual gaps are back- then forward-filled.
4f — Lagged-load exog (opt-in, default OFF, gated by config.include_load_lag_exog), via preprocessing.load_lags.build_load_lag_features. Runs AFTER the Step-5 backfill so a load-lag coverage failure is never masked by it. Appends load_lag_<L> / load_<zone>_lag_<L> / share_<zone>_lag_<L> columns per config.load_lag_hours / config.load_lag_sources. A coverage or staleness failure (or a missing zone-interim file for the "zones"/"zone_shares" sources) is governed by config.on_load_lag_failure: "raise" re-raises; "skip" logs a warning and omits the columns (fail-safe). Provider-based exogenous columns are then appended via build_providers_from_config (requires spotforecast2-safe >= 15.7.0). The active providers are governed by the config flags include_covid_infection_rate, include_entsoe_forecast_load, include_entsoe_renewable_forecast, include_entsoe_net_load, and include_entsoe_day_ahead_price. Cyclical (sine/cosine) encoding is then applied via apply_cyclical_encoding, and degree-config.poly_features_degree interaction terms are added via create_interaction_features. When the degree is at least 2, the polynomial columns are ranked by mutual information with the primary target and capped to config.max_poly_features via select_top_poly_features.
6 — The training feature set is chosen via select_exogenous_features (including, when config.include_day_type_features is set, the day-type columns is_workday / day_type), with provider and load-lag columns appended (order-preserving, de-duplicated).
7 — Targets and covariates are merged via merge_data_and_covariates into self.data_with_exog and the forecast-horizon covariates self.exo_pred.

When config.use_exogenous_features is False the method is a no-op and returns self immediately, leaving the pipeline target-only.

Attributes

Name	Type	Description
weather_aligned	pd.DataFrame	Weather frame aligned to the pipeline index, reused by the interaction and selection steps.
zone_weather_aligned	Dict[str, pd.DataFrame]	Per-zone weather frames keyed by target name, indexed over `[data_start, cov_end]` (covering the forecast horizon). Populated only when `config.per_zone_weather` is True and every zone fetch succeeded; empty otherwise (including the fail-safe “skip” degradation). Consumed at the per-target seam in `_get_target_data` to overwrite the shared weather columns.
exogenous_features	pd.DataFrame	Full combined, encoded, and capped exogenous feature matrix.
exog_feature_names	List[str]	Names of the exogenous features selected for training (including provider and load-lag columns).
data_with_exog	pd.DataFrame	Target data merged with the selected exogenous covariates.
exo_pred	pd.DataFrame	Exogenous covariates spanning the forecast horizon, supplied to the forecaster at predict time.

Returns

Name	Type	Description
	BaseTask	`self` (for method chaining).

Raises

Name	Type	Description
	RuntimeError	If `prepare_data` has not been called.
	`WeatherFetchError`	If the Open-Meteo fetch fails (single-point, population-weighted, per-zone, or `zone_weather_columns` path) and `config.on_weather_failure == "raise"`.
	`LoadLagError`	If `config.include_load_lag_exog` is set, the load-lag builder cannot produce NaN-free columns (stale source, excessive staleness), and `config.on_load_lag_failure == "raise"`.
	FileNotFoundError	If `config.include_load_lag_exog` is set with `load_lag_sources` in `{"zones", "zone_shares"}`, the zone-interim CSV is missing, and `config.on_load_lag_failure == "raise"`.

Examples

With exogenous features disabled the method is a no-op, so the example below runs without any network access and leaves the pipeline target-only.

import tempfile
import pandas as pd
import numpy as np
from spotforecast2_safe.multitask import MultiTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

rng = np.random.default_rng(0)
idx = pd.date_range("2023-01-01", periods=24 * 14, freq="h", tz="UTC")
df = pd.DataFrame({"a": rng.normal(100, 10, len(idx))}, index=idx)
df.index.name = "DateTime"

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(
        predict_size=6,
        use_exogenous_features=False,
        use_outlier_detection=False,
        cache_home=tmp,
    )
    mt = MultiTask(cfg, dataframe=df)
    mt.prepare_data().detect_outliers().impute().build_exogenous_features()
    print(f"Exogenous features used: {mt.config.use_exogenous_features}")
    print(f"Selected exog feature names: {mt.exog_feature_names}")

Exogenous features used: False
Selected exog feature names: []

create_forecaster

multitask.predict.PredictTask.create_forecaster(target=None)

Create a fresh forecaster for the given target.

Delegates to config.forecaster_factory when set; otherwise falls back to default_lgbm_forecaster_factory. This factory hook lets callers swap the estimator without subclassing BaseTask.

Parameters

Name	Type	Description	Default
target	Optional[str]	Optional target column name. Forwarded to the factory so that custom factories can specialise per target.	`None`

Returns

Name	Type	Description
	Any	A new, unfitted forecaster instance.

Examples

import tempfile
from pathlib import Path
from spotforecast2_safe.multitask import LazyTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(
        predict_size=6,
        use_exogenous_features=False,
        cache_home=Path(tmp),
    )
    task = LazyTask(cfg)
    forecaster = task.create_forecaster()
print(f"Type: {type(forecaster).__name__}")
print(f"Lags: {forecaster.lags}")

Type: ForecasterRecursive
Lags: [ 1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23]

cv_ts

multitask.predict.PredictTask.cv_ts(y_train)

Build a TimeSeriesFold for cross-validation.

Constructs the cross-validation splitter used by all tuning tasks. Internally uses sklearn.model_selection.TimeSeriesSplit to compute split boundaries that respect temporal ordering and avoid data leakage between folds.

The validation boundary is determined by run_state.end_train_ts minus config.delta_val. When config.train_size is set, the sklearn splitter uses a sliding fixed-size training window (max_train_size); otherwise an expanding window is used.

Parameters

Name	Type	Description	Default
y_train	pd.Series	Training time series for the current target. Used both to determine the validation boundary and as the sequence passed to `TimeSeriesSplit.split` to derive `initial_train_size`.	required

Returns

Name	Type	Description
	TimeSeriesFold	A configured `TimeSeriesFold` instance ready to be passed to
	TimeSeriesFold	a model-selection function.

Examples

import tempfile
import numpy as np
import pandas as pd
from spotforecast2_safe.multitask import MultiTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

rng = np.random.default_rng(0)
idx = pd.date_range("2023-01-01", periods=24 * 14, freq="h", tz="UTC")
df = pd.DataFrame({"a": rng.normal(100, 10, len(idx))}, index=idx)
df.index.name = "DateTime"

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(
        predict_size=6,
        use_exogenous_features=False,
        use_outlier_detection=False,
        cache_home=tmp,
        number_folds=2,
        auto_save_models=False,
        verbose=False,
    )
    mt = MultiTask(cfg, dataframe=df)
    mt.prepare_data().detect_outliers().impute().build_exogenous_features()
    y_train = mt.df_pipeline["a"]
    cv = mt.cv_ts(y_train)
    print(f"TimeSeriesFold steps: {cv.steps}")
    print(f"initial_train_size: {cv.initial_train_size}")

TimeSeriesFold steps: 6
initial_train_size: 324

detect_outliers

multitask.predict.PredictTask.detect_outliers()

Apply hard-bound filtering and IsolationForest outlier detection.

Hard bounds from config.bounds are applied to the pipeline data (out-of-bound values are removed and later filled by impute()). IsolationForest detection (config.use_outlier_detection) is advisory: detected outliers are logged per column but not removed.

Returns

Name	Type	Description
	BaseTask	`self` (for method chaining).

Raises

Name	Type	Description
	RuntimeError	If method `prepare_data` has not been called.

Examples

import tempfile
import numpy as np
import pandas as pd
from spotforecast2_safe.multitask import MultiTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

rng = np.random.default_rng(0)
idx = pd.date_range("2023-01-01", periods=24 * 14, freq="h", tz="UTC")
df = pd.DataFrame({"a": rng.normal(100, 10, len(idx))}, index=idx)
df.index.name = "DateTime"

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(
        predict_size=6,
        use_exogenous_features=False,
        use_outlier_detection=False,
        cache_home=tmp,
        auto_save_models=False,
        verbose=False,
    )
    mt = MultiTask(cfg, dataframe=df)
    mt.prepare_data()
    mt.detect_outliers()
    print(f"Pipeline shape: {mt.df_pipeline.shape}")
    assert mt.df_pipeline_original is not None

Pipeline shape: (336, 1)

impute

multitask.predict.PredictTask.impute()

Fill missing values using the configured imputation strategy.

Returns

Name	Type	Description
	BaseTask	`self` (for method chaining).

Raises

Name	Type	Description
	RuntimeError	If method `prepare_data` has not been called.

Examples

import tempfile
import numpy as np
import pandas as pd
from spotforecast2_safe.multitask import MultiTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

rng = np.random.default_rng(0)
idx = pd.date_range("2023-01-01", periods=24 * 14, freq="h", tz="UTC")
values = rng.normal(100, 10, len(idx))
values[10:13] = float("nan")  # inject a few gaps
df = pd.DataFrame({"a": values}, index=idx)
df.index.name = "DateTime"

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(
        predict_size=6,
        use_exogenous_features=False,
        use_outlier_detection=False,
        cache_home=tmp,
        auto_save_models=False,
        verbose=False,
    )
    mt = MultiTask(cfg, dataframe=df)
    mt.prepare_data().detect_outliers().impute()
    missing = mt.df_pipeline["a"].isna().sum()
    print(f"Missing values after imputation: {missing}")
    assert missing == 0

Missing values after imputation: 0

load_models

multitask.predict.PredictTask.load_models(
    task_name=None,
    target=None,
    max_age_days=None,
)

Load the most recent fitted models from the cache directory.

Scans <cache_home>/models/<data_frame_name>/ for .joblib files matching the current data_frame_name. Optionally filters by task_name, target, and max_age_days.

Parameters

Name	Type	Description	Default
task_name	Optional[str]	If given, only load models from this task (`"lazy"`, `"defaults"`, `"optuna"`, or `"spotoptim"`). `None` accepts any task.	`None`
target	Optional[str]	If given, only load the model for this target column. `None` loads the most recent model for every target found.	`None`
max_age_days	Optional[float]	Maximum age in days. Models older than this are ignored. `None` accepts any age.	`None`

Returns

Name	Type	Description
	Dict[str, Any]	Mapping `{target: forecaster}` of loaded model objects.
	Dict[str, Any]	Empty dict if no matching models were found.

Examples

import tempfile
from pathlib import Path
from spotforecast2_safe.multitask import LazyTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(
        data_frame_name="demo",
        cache_home=Path(tmp),
        verbose=False,
    )
    task = LazyTask(cfg)
    # Save a dummy object, then load it back.
    dummy_forecaster = {"lags": [1, 2, 24]}
    task.save_models(
        task_name="lazy",
        forecasters={"load": dummy_forecaster},
    )
    loaded = task.load_models(task_name="lazy")
    print(f"Loaded targets: {list(loaded.keys())}")
    assert loaded["load"]["lags"] == [1, 2, 24]

Loaded targets: ['load']

load_tuning_results

multitask.predict.PredictTask.load_tuning_results(
    target,
    task_name=None,
    max_age_days=None,
)

Load the most recent tuning results for a target from cache.

Scans <cache_home>/tuning_results/ for files matching the current data_frame_name and target. Optionally filters by task_name and discards results older than max_age_days.

Parameters

Name	Type	Description	Default
target	str	Name of the forecast target column.	required
task_name	Optional[str]	If given, only consider results from this tuning algorithm (e.g. `"optuna"` or `"spotoptim"`). `None` accepts any algorithm.	`None`
max_age_days	Optional[float]	Maximum age in days. Results older than this are ignored. `None` accepts any age.	`None`

Returns

Name	Type	Description
	Optional[Dict[str, Any]]	A dictionary with keys `best_params`, `best_lags`,
	Optional[Dict[str, Any]]	`task_name`, `target`, `data_frame_name`, and
	Optional[Dict[str, Any]]	`timestamp`; or `None` if no matching file was found.

Examples

import tempfile
from pathlib import Path
from spotforecast2_safe.multitask import LazyTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(data_frame_name="demo10", cache_home=Path(tmp))
    task = LazyTask(cfg)
    task.save_tuning_results(
        target="target_0",
        task_name="optuna",
        best_params={"n_estimators": 100},
        best_lags=24,
    )
    result = task.load_tuning_results(target="target_0")
    print(result["best_params"])

{'n_estimators': 100}

log_summary

multitask.predict.PredictTask.log_summary()

Log a summary of the current pipeline configuration.

Examples

import tempfile
import numpy as np
import pandas as pd
from spotforecast2_safe.multitask import MultiTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

rng = np.random.default_rng(0)
idx = pd.date_range("2023-01-01", periods=24 * 14, freq="h", tz="UTC")
df = pd.DataFrame({"a": rng.normal(100, 10, len(idx))}, index=idx)
df.index.name = "DateTime"

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(
        predict_size=6,
        use_exogenous_features=False,
        use_outlier_detection=False,
        cache_home=tmp,
        auto_save_models=False,
        verbose=False,
    )
    mt = MultiTask(cfg, dataframe=df)
    mt.prepare_data().detect_outliers().impute().build_exogenous_features()
    # log_summary writes to the pipeline logger; call it to confirm
    # it runs without error.
    mt.log_summary()
    print("log_summary completed without error")

log_summary completed without error

plot_with_outliers

multitask.predict.PredictTask.plot_with_outliers()

Visualise original vs. cleaned data with outlier markers.

Raises

Name	Type	Description
	RuntimeError	If method `detect_outliers` has not been called.
	NotImplementedError	Always — plotting is not available in `spotforecast2-safe`. Use the `spotforecast2` package for visualisation.

Examples

import tempfile
import numpy as np
import pandas as pd
from spotforecast2_safe.multitask import MultiTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

rng = np.random.default_rng(0)
idx = pd.date_range("2023-01-01", periods=24 * 14, freq="h", tz="UTC")
df = pd.DataFrame({"a": rng.normal(100, 10, len(idx))}, index=idx)
df.index.name = "DateTime"

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(
        predict_size=6,
        use_exogenous_features=False,
        use_outlier_detection=False,
        cache_home=tmp,
        auto_save_models=False,
        verbose=False,
    )
    mt = MultiTask(cfg, dataframe=df)
    mt.prepare_data().detect_outliers()
    try:
        mt.plot_with_outliers()
    except NotImplementedError as exc:
        print(f"Plotting unavailable in spotforecast2-safe: {exc}")

Plotting unavailable in spotforecast2-safe: Plotting is not available in spotforecast2-safe (no plotly/matplotlib). Use the spotforecast2 package for visualisation.

prepare_data

multitask.predict.PredictTask.prepare_data(demo_data=None, df_test=None)

Load, resample, validate, and configure the pipeline data.

Uses the following precedence for the training data:

demo_data argument (if provided).
self._dataframe set via the constructor.

Similarly for test data:

df_test argument (if provided).
self.data_test set via the constructor.
self.config.test_data_loader(self.config) if set.

Parameters

Name	Type	Description	Default
demo_data	Optional[pd.DataFrame]	Pre-loaded input DataFrame. When `None`, the constructor `dataframe` is used.	`None`
df_test	Optional[pd.DataFrame]	Pre-loaded test DataFrame. When `None`, the constructor `data_test` is used, then `config.test_data_loader`.	`None`

Returns

Name	Type	Description
	BaseTask	`self` (for method chaining).

Raises

Name	Type	Description
	ValueError	If no data source is available (no `demo_data`, no constructor `dataframe`).

Examples

import tempfile
import pandas as pd
import numpy as np
from spotforecast2_safe.multitask import MultiTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

rng = np.random.default_rng(0)
idx = pd.date_range("2023-01-01", periods=24 * 14, freq="h", tz="UTC")
df = pd.DataFrame({"a": rng.normal(100, 10, len(idx))}, index=idx)
df.index.name = "DateTime"

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(
        predict_size=6,
        use_exogenous_features=False,
        use_outlier_detection=False,
        cache_home=tmp,
    )
    mt = MultiTask(cfg, dataframe=df)
    mt.prepare_data()
    print(f"Pipeline shape: {mt.df_pipeline.shape}")
    print(f"Targets: {mt.run_state.targets}")

Pipeline shape: (336, 1)
Targets: ['a']

run

multitask.predict.PredictTask.run(
    show=False,
    task_name=None,
    max_age_days=None,
    **kwargs,
)

Run prediction using previously saved models.

Parameters

Name	Type	Description	Default
show	bool	If `True`, invoke the visualisation hooks.	`False`
task_name	Optional[str]	Restrict model loading to a specific source task (`"lazy"`, `"defaults"`, `"optuna"`, or `"spotoptim"`). `None` loads the most recent model regardless of source.	`None`
max_age_days	Optional[float]	Maximum age in days for saved models. Models older than this are ignored. `None` accepts any age.	`None`

Returns

Name	Type	Description
	Dict[str, Any]	Aggregated prediction package. Per-target packages are stored
	Dict[str, Any]	on `self.results["predict"]`.

Raises

Name	Type	Description
	RuntimeError	If no saved models are found in the cache directory, or if a target has no matching model.

Examples

import tempfile
import warnings
from pathlib import Path

import numpy as np
import pandas as pd

from spotforecast2_safe.configurator.config_multi import ConfigMulti
from spotforecast2_safe.multitask.lazy import LazyTask
from spotforecast2_safe.multitask.predict import PredictTask

warnings.filterwarnings("ignore")

rng = np.random.default_rng(0)
idx = pd.date_range("2023-01-01", periods=300, freq="h", tz="UTC")
df = pd.DataFrame(
    {"total_load_actual": rng.normal(10_000, 500, 300).clip(8_000, 12_000)},
    index=idx,
)
df.index.name = "DateTime"

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(
        data_frame_name="demo",
        predict_size=12,
        auto_save_models=True,
        cache_home=Path(tmp),
        targets=["total_load_actual"],
        verbose=False,
        use_outlier_detection=False,
        use_exogenous_features=False,
        number_folds=2,
    )
    # Train and save a model with LazyTask first.
    lazy = LazyTask(cfg, dataframe=df)
    lazy.prepare_data()
    lazy.run()

    # Run PredictTask to load the saved model and forecast.
    task = PredictTask(cfg, dataframe=df)
    task.prepare_data()
    result = task.run()

    print(f"Future predictions shape: {result['future_pred'].shape}")
    assert result["future_pred"].shape == (12,)
    assert "metrics_train" in result

Future predictions shape: (12,)

save_models

multitask.predict.PredictTask.save_models(task_name, forecasters=None)

Save fitted forecaster models to the cache directory.

Each model is serialised with joblib (compress=3) into <cache_home>/models/<data_frame_name>/ using a datetime-stamped filename so that multiple snapshots can coexist.

Filename format::

<data_frame_name>_<target>_<task_name>_<YYYYMMDD_HHMMSS>.joblib

If forecasters is None the method collects fitted models from self.results[task_name], where each prediction package is expected to contain a "forecaster" key.

Parameters

Name	Type	Description	Default
task_name	str	Task identifier (`"lazy"`, `"defaults"`). The names `"optuna"` and `"spotoptim"` are also accepted so that model caches produced by the `spotforecast2` sibling package can be saved and loaded; no tuning is performed in this package.	required
forecasters	Optional[Dict[str, Any]]	Optional mapping `{target: fitted_forecaster}`. When `None`, models are taken from the prediction packages stored in `self.results`.	`None`

Returns

Name	Type	Description
	Dict[str, Path]	Mapping `{target: Path}` of saved model file paths.

Raises

Name	Type	Description
	ValueError	If `task_name` is not one of `"lazy"`, `"defaults"`, `"optuna"`, `"spotoptim"`.
	RuntimeError	If no fitted models are available for the requested task.

Examples

import tempfile
from pathlib import Path
from spotforecast2_safe.multitask import LazyTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(
        data_frame_name="demo",
        cache_home=Path(tmp),
        verbose=False,
    )
    task = LazyTask(cfg)
    # Supply a tiny in-memory object as a stand-in for a fitted forecaster.
    dummy_forecaster = object()
    saved = task.save_models(
        task_name="lazy",
        forecasters={"load": dummy_forecaster},
    )
    print(f"Saved targets: {list(saved.keys())}")
    assert saved["load"].suffix == ".joblib"

Saved targets: ['load']

save_tuning_results

multitask.predict.PredictTask.save_tuning_results(
    target,
    task_name,
    best_params,
    best_lags,
)

Save tuning results (best parameters and lags) to a JSON file.

The file is stored under <cache_home>/tuning_results/ with a datetime-stamped filename so that loaders can determine freshness.

Filename format::

<data_frame_name>_<target>_<task_name>_<YYYYMMDD_HHMMSS>.json

Parameters

Name	Type	Description	Default
target	str	Name of the forecast target column.	required
task_name	str	Tuning algorithm identifier (e.g. `"optuna"`, `"spotoptim"`).	required
best_params	Dict[str, Any]	Best hyperparameters discovered during tuning.	required
best_lags	Any	Best lag configuration (int, list, or nested list).	required

Returns

Name	Type	Description
	Path	Path to the saved JSON file.

Examples

import tempfile
from pathlib import Path
from spotforecast2_safe.multitask import LazyTask
from spotforecast2_safe.configurator.config_multi import ConfigMulti

with tempfile.TemporaryDirectory() as tmp:
    cfg = ConfigMulti(data_frame_name="demo10", cache_home=Path(tmp))
    task = LazyTask(cfg)
    path = task.save_tuning_results(
        target="target_0",
        task_name="optuna",
        best_params={"n_estimators": 100, "learning_rate": 0.05},
        best_lags=[1, 2, 24],
    )
    print(path.name[:10])

demo10_tar