Data¶

Setup-time facade for parquet-backed time series. For the architecture-level role, see Architecture → Data pipeline. Nothing in powerzoojax.data enters the compiled rollout.

Prefer load_jax_profiles(...) for env construction and load_signals(...) for inspection. load_data(...) covers raw-column layouts and direct table access; benchmark code standardizes on load_jax_profiles(...) and the parquet facade documented in Architecture → Data pipeline.

Semantic loading¶

from powerzoojax.data import DataLoader, signals as S

loader = DataLoader()
profiles = loader.load_jax_profiles(
    [S.LOAD_ACTUAL_MW, S.SOLAR_AVAILABLE_MW],
    source="gb",
    start_date="2025-04-01",
    end_date="2025-12-31",
    resample="30min",
)

Stable signal names¶

Power-system signals:

LOAD_ACTUAL_MW
LOAD_FORECAST_DA_MW
LOAD_FORECAST_P10_MW
LOAD_FORECAST_P50_MW
LOAD_FORECAST_P90_MW
SOLAR_AVAILABLE_MW
WIND_AVAILABLE_MW

Market index signals:

MARKET_MID_PRICE_APX
MARKET_MID_PRICE_N2EX
MARKET_MID_VOLUME_APX
MARKET_MID_VOLUME_N2EX

Data-center signals:

DC_CPU_UTIL
DC_MEM_UTIL
DC_NET_IN
DC_NET_OUT
DC_DISK_IO
DC_POWER_MW
DC_GPU_UTIL
DC_GPU_MEM_UTIL

Loader, manifest, registry¶

`powerzoojax.data.data_loader.DataLoader(data_dir=None, manifest_dir=None)` ¶

Benchmark data facade backed by parquet files and manifest metadata.

Features: - JAX array output via :meth:load_jax_profiles (float32, GPU-ready) - Semantic signal loading via :meth:load_signals - Calendar / profile time-alignment across heterogeneous sources - Forecast-panel loading with issue_time / target_time - Legacy column-based loading via :meth:load_data - Resampling, date filtering, multi-dataset merge

Initialize DataLoader.

Parameters:

Name	Type	Description	Default
`data_dir`	`Optional[Union[str, Path]]`	Directory containing parquet files. If None, defaults to `powerzoojax/data/parquet`.	`None`
`manifest_dir`	`Optional[Union[str, Path]]`	Directory containing manifest JSON files. If None, defaults to `powerzoojax/data/manifests`.	`None`

`registry` `property` ¶

Expose the dataset registry for advanced querying.

`load_jax_profiles(signals, *, source=None, region=None, start_date=None, end_date=None, resample=None, time_alignment=None, interpolation='linear', dtype=jnp.float32)` ¶

Load semantic signals and return as a JAX array.

This is the primary entry-point for env construction. It calls :meth:load_signals internally, then converts numeric columns to a jnp.ndarray of shape (T, n_signals).

The result is a JAX array on the default device (GPU when available), ready to be passed to make_trans_params(case, load_profiles=...).

Parameters:

Name	Type	Description	Default
`signals`	`List[str]`	Semantic signal names (e.g. `["load.actual_mw", "solar.available_mw"]`).	required
`source`	`Optional[str]`	Restrict lookup to a specific data source.	`None`
`region`	`Optional[str]`	Filter by region (e.g. `"NSW1"`).	`None`
`start_date`	`Optional[Union[str, datetime, date]]`	Simulation time window start (inclusive).	`None`
`end_date`	`Optional[Union[str, datetime, date]]`	Simulation time window end (inclusive).	`None`
`resample`	`Optional[str]`	Target frequency (`"5min"`, `"30min"`, ...).	`None`
`time_alignment`	`Optional[Dict[str, str]]`	Per-signal calendar-shift overrides.	`None`
`interpolation`	`str`	Interpolation method when resampling.	`'linear'`
`dtype`	`dtype`	JAX dtype for the output array. Default `jnp.float32`.	`float32`

Returns:

Type	Description
`ndarray`	`jnp.ndarray` of shape `(T, n_signals)` with columns
`ndarray`	ordered as signals.

`load_jax_array(df, columns=None, dtype=jnp.float32)` ¶

Convert a DataFrame (e.g. from :meth:load_signals) to JAX.

Parameters:

Name	Type	Description	Default
`df`	`DataFrame`	Source DataFrame.	required
`columns`	`Optional[List[str]]`	Columns to extract (order preserved). If None, all numeric columns are used.	`None`
`dtype`	`dtype`	Target JAX dtype.	`float32`

Returns:

Type	Description
`ndarray`	`jnp.ndarray` of shape `(n_rows, n_columns)`.

`load_signals(signals, *, source=None, region=None, start_date=None, end_date=None, resample=None, time_alignment=None, interpolation='linear')` ¶

Load one or more semantic signals into a single DataFrame.

Parameters:

Name	Type	Description	Default
`signals`	`List[str]`	Semantic signal names (e.g. `["load.actual_mw", "solar.available_mw"]`).	required
`source`	`Optional[str]`	Restrict lookup to a specific data source (`"gb"`, `"aemo"`, `"alibaba"`, ...).	`None`
`region`	`Optional[str]`	Filter by region (e.g. `"NSW1"`).	`None`
`start_date`	`Optional[Union[str, datetime, date]]`	Simulation time window start (inclusive).	`None`
`end_date`	`Optional[Union[str, datetime, date]]`	Simulation time window end (inclusive).	`None`
`resample`	`Optional[str]`	Target frequency (`"5min"`, `"30min"`, ...).	`None`
`time_alignment`	`Optional[Dict[str, str]]`	Per-signal calendar-shift overrides.	`None`
`interpolation`	`str`	Interpolation method when resampling.	`'linear'`

Returns:

Type	Description
`DataFrame`	DataFrame indexed by `datetime` (simulation timeline).

`load_actual_series(signals, **kwargs)` ¶

Convenience wrapper: load only actual_series data.

`load_forecast_panel(signals, *, source=None, region=None, start_date=None, end_date=None)` ¶

Load forecast-panel data with issue_time + target_time.

`list_available_datasets()` ¶

List all available datasets (parquet files).

`get_metadata(dataset_name)` ¶

Get metadata for a dataset.

`get_available_columns(dataset_name)` ¶

Get available columns for a dataset.

`get_date_range(dataset_name)` ¶

Get date range for a dataset.

`find_datasets_for_columns(columns)` ¶

Find which datasets contain the requested columns.

`load_data(dataset_name=None, columns=None, start_date=None, end_date=None, validate_dates=True, merge_how='inner', resample=None, interpolation='linear')` ¶

Load data from parquet file(s) with optional resampling.

If dataset_name is not provided, automatically finds datasets based on columns.

`print_dataset_info(dataset_name)` ¶

Print information about a dataset.

`get_summary()` ¶

Get summary of currently loaded data.

`powerzoojax.data.manifest.DatasetManifest(name, source, data_type, time_mode, resolution, parquet_file, column_map=dict(), index_map=dict(), derived=dict(), normalize=dict(), data_epoch=None, cyclical=False, region_values=list(), date_range=None, metadata_json=None, source_url=None, source_urls=None, source_organization=None)` `dataclass` ¶

Declarative description of one dataset.

A manifest is the single source of truth that tells the data layer: * which parquet file to read, * how to map raw columns to semantic signals, * how to handle time alignment (calendar vs. profile), and * whether data can be tiled cyclically.

`signals` `property` ¶

All semantic signals this manifest can provide (mapped + derived).

`raw_columns_needed` `property` ¶

Raw columns that must be read from the parquet file.

`from_json(path)` `classmethod` ¶

Load a manifest from a JSON file.

`to_dict()` ¶

Serialize to a plain dict (JSON-friendly).

`powerzoojax.data.registry.DatasetRegistry(manifest_dir=None)` ¶

Load all manifest JSONs and build a reverse index from signal -> datasets.

`find_by_signal(signal, source=None, data_type=None)` ¶

Return manifests that provide signal, optionally filtered.

`resolve_signals(signals, source=None)` ¶

Map each requested signal to the best matching manifest.

Returns {signal: manifest}. Raises if any signal cannot be resolved.

`powerzoojax.data.alignment.TimeAligner` ¶

Map raw-data timestamps onto a unified simulation timeline.

`align_calendar(df, sim_start, sim_end, align_from=None, time_col='datetime')` `staticmethod` ¶

Shift a calendar-mode dataset onto [sim_start, sim_end].

`align_profile(df, sim_start, sim_end, resolution, time_col='datetime')` `staticmethod` ¶

Tile a profile cyclically to cover [sim_start, sim_end].

Uses jnp.tile so the tiled numeric data stays on the JAX default device (GPU when available).

`tile_profile_jax(values, n_needed)` `staticmethod` ¶

Tile a profile array to length n_needed using jnp.tile.

Parameters¶

values : (n_profile, n_signals) JAX array. n_needed : target number of time steps.

Returns¶

(n_needed, n_signals) JAX array, dtype float32.

`align(df, *, time_mode, sim_start, sim_end, align_from=None, resolution='30min', time_col='datetime')` `classmethod` ¶

Dispatch to the correct alignment strategy.

Frozen splits¶

from powerzoojax.data import (
    GB_TRAIN_START, GB_TRAIN_END,
    GB_IID_START, GB_IID_END,
    AUSGRID_TRAIN_START, AUSGRID_TRAIN_END,
    AUSGRID_IID_START, AUSGRID_IID_END,
    AUSGRID_SUMMER_START, AUSGRID_SUMMER_END,
    gb_windows,
    ausgrid_windows,
)

`powerzoojax.data.splits.gb_windows()` ¶

Return ((train_start, train_end), (iid_start, iid_end)).

`powerzoojax.data.splits.ausgrid_windows()` ¶

Return (train_window, iid_window, summer_ood_window).

Ausgrid utilities¶

from powerzoojax.data import (
    AUSGRID_FEEDER_POOLS,
    AUSGRID_TRAIN_POOL,
    AUSGRID_ZONE_HOLDOUT_POOL,
    get_ausgrid_split,
    get_feeder_substations,
    select_full_coverage_substations,
)

`powerzoojax.data.ausgrid_utils.get_ausgrid_split(role)` ¶

Return (start, end, substation_list) for a named split role.

Parameters:

Name	Type	Description	Default
`role`	`str`	One of `"train"`, `"iid"`, `"summer_ood"`, `"zone_holdout"`.	required

Returns:

Type	Description
`Tuple[str, str, List[str]]`	Tuple of (ISO-date start, ISO-date end, list of substation names).

Raises:

Type	Description
`KeyError`	if role is not recognised.

`powerzoojax.data.ausgrid_utils.get_feeder_substations(feeder, role='train')` ¶

Return substation list for a specific feeder and role.

Parameters:

Name	Type	Description	Default
`feeder`	`str`	`"feeder_A"`, `"feeder_B"`, or `"feeder_C"`.	required
`role`	`str`	`"train"` or `"zone_holdout"`.	`'train'`

Returns:

Type	Description
`List[str]`	List of substation names for that feeder/role combination.

`powerzoojax.data.ausgrid_utils.select_full_coverage_substations(df, start, end, *, max_gap_steps=2, substation_col='region', time_col='datetime', resolution='15min')` ¶

Return substations that have no consecutive time gap > max_gap_steps.

A gap is defined as a run of consecutive missing 15-min slots within [start, end]. Substations with any gap run longer than max_gap_steps are excluded.

Parameters:

Name	Type	Description	Default
`df`	`DataFrame`	DataFrame with at least `time_col` and `substation_col` columns. Multi-region format (one row per region per timestamp).	required
`start`	`str`	ISO date string, inclusive window start.	required
`end`	`str`	ISO date string, inclusive window end.	required
`max_gap_steps`	`int`	Maximum tolerated consecutive missing steps.	`2`
`substation_col`	`str`	Column name identifying the substation/region.	`'region'`
`time_col`	`str`	Column name for timestamps.	`'datetime'`
`resolution`	`str`	Expected time resolution for the full grid.	`'15min'`

Returns:

Type	Description
`List[str]`	Sorted list of substation names with full (or near-full) coverage.

Non-stationary sampling¶

from powerzoojax.data import EpisodeConfig, NonstationarySampler, apply_drift

`powerzoojax.data.nonstationary.EpisodeConfig` ¶

Bases: NamedTuple

Episode-level non-stationarity parameters.

`powerzoojax.data.nonstationary.NonstationarySampler(total_steps, total_episodes=500, *, max_steps=48, drift_rate=0.0003, rolling_sigma_steps=720.0)` ¶

Generates non-stationary episode parameters for DSO task.

Usage::

sampler = NonstationarySampler(
    total_steps=10080,     # 7 months × 48 steps/day
    total_episodes=500,
)
cfg = sampler.sample(episode_idx=100, key=jax.random.PRNGKey(0))
# → EpisodeConfig(episode_start=2340, drift_factor=1.03)

The sampler supports four modes:

"train" — rolling center + drift (default)
"iid" — uniform random start, no drift (same window)
"drift_shock" — fixed high drift factor (OOD)
"fixed" — fixed start, no drift (for reproducible eval)

Parameters:

Name	Type	Description	Default
`total_steps`	`int`	Length of the full feeder shape array (in time steps). For Ausgrid training window (7 months @ 48 steps/day) ≈ 10,080.	required
`total_episodes`	`int`	Expected total training episodes (for phase_k calc).	`500`
`max_steps`	`int`	Episode length (default 48 = 24h @ 30min).	`48`
`drift_rate`	`float`	Per-episode drift rate (default 0.0003 = +0.03%/episode).	`0.0003`
`rolling_sigma_steps`	`float`	Gaussian σ for month-rolling start sampling, in time steps (default 720 = 15 days × 48 steps/day).	`720.0`

`sample(episode_idx, key, *, mode='train')` ¶

Sample episode parameters.

Parameters:

Name	Type	Description	Default
`episode_idx`	`int`	Current episode index (0-based).	required
`key`	`Array`	JAX PRNG key.	required
`mode`	`str`	Sampling mode — `"train"`, `"iid"`, `"drift_shock"`, or `"fixed"`.	`'train'`

Returns:

Type	Description
`EpisodeConfig`	`EpisodeConfig(episode_start, drift_factor)`.

`sample_batch(start_idx, batch_size, key, *, mode='train')` ¶

Sample a batch of episode configs.

Parameters:

Name	Type	Description	Default
`start_idx`	`int`	First episode index in the batch.	required
`batch_size`	`int`	Number of episodes.	required
`key`	`Array`	JAX PRNG key (split internally).	required
`mode`	`str`	Sampling mode.	`'train'`

Returns:

Type	Description
`list`	List of `EpisodeConfig` of length `batch_size`.

`powerzoojax.data.nonstationary.apply_drift(load_profiles_p, load_profiles_q, drift_factor)` ¶

Scale load profiles by a drift factor.

Parameters:

Name	Type	Description	Default
`load_profiles_p`	`ndarray`	(T, n_bus) active load in p.u.	required
`load_profiles_q`	`ndarray`	(T, n_bus) reactive load in p.u.	required
`drift_factor`	`float`	Multiplicative scaling (e.g. 1.03 for +3% load growth).	required

Returns:

Type	Description
`Tuple[ndarray, ndarray]`	Scaled `(load_profiles_p, load_profiles_q)`.

DC microgrid profiles¶

from powerzoojax.data import (
    cycle_profile,
    make_synthetic_cpu_profile,
    make_synthetic_solar_profile,
    make_synthetic_outdoor_temp_profile,
    make_all_synthetic_profiles,
    load_workload_profiles,
    apply_ood_transform,
    VALID_SOURCES,
    VALID_OOD_SCENARIOS,
)

apply_ood_transform(params, scenario) produces the OOD splits used by the DC Microgrid benchmark. Valid scenarios: workload_swap, workload_shock, renewable_drought, cooling_stress, dg_derating, sla_tighten.

`powerzoojax.data.dc_microgrid_profiles.cycle_profile(arr, episode_len, start_step=0)` ¶

Slice or cyclically tile arr to exactly episode_len steps.

Parameters:

Name	Type	Description	Default
`arr`	`ndarray`	Source array of shape `(T,)` (float32). If empty (T=0), returned as-is.	required
`episode_len`	`int`	Desired output length.	required
`start_step`	`int`	Starting offset into the cyclically-tiled sequence.	`0`

Returns:

Type	Description
`ndarray`	Array of shape `(episode_len,)` with dtype preserved.

Rules

If T >= episode_len and start_step == 0: simple slice.
Otherwise: tile enough repetitions to cover start_step + episode_len, then slice.

`powerzoojax.data.dc_microgrid_profiles.make_all_synthetic_profiles(episode_len=DC_EPISODE_LEN, steps_per_day=DC_STEPS_PER_DAY)` ¶

Return all three synthetic profiles as a dict.

`powerzoojax.data.dc_microgrid_profiles.load_workload_profiles(source='google', episode_len=DC_EPISODE_LEN, start_step=0, data_dir=None, manifest_dir=None, strict=True, require_real_data=False)` ¶

Load DC workload profiles, with controlled fallback behaviour.

Parameters:

Name	Type	Description	Default
`source`	`str`	Data source key — `"google"`, `"azure"`, or `"alibaba"`.	`'google'`
`episode_len`	`int`	Number of steps in the output profiles (default 288).	`DC_EPISODE_LEN`
`start_step`	`int`	Cyclic offset into the source profile (default 0).	`0`
`data_dir`	`Optional[str]`	Override for the parquet directory (`None` → default).	`None`
`manifest_dir`	`Optional[str]`	Override for the manifests directory (`None` → default).	`None`
`strict`	`bool`	Schema / manifest error handling (default `True`). `strict=True` (default): non-infra exceptions (schema mismatch, signal-resolution failure, …) are re-raised immediately. `strict=False`: all exceptions warn + synthetic fallback.	`True`
`require_real_data`	`bool`	Infrastructure error handling (default `False`). `False` (default): `FileNotFoundError` and `ImportError` (parquet engine absent) → `UserWarning` + synthetic fallback. Suitable for generic env construction and development. `True`: `FileNotFoundError` and `ImportError` are ALSO re-raised — the caller requires a real workload profile. Use for OOD scenarios (`workload_swap`, `workload_shock`) where using synthetic data would silently corrupt the experiment.	`False`

Returns:

Type	Description
`Dict[str, ndarray]`	Dict with keys `"cpu_profile"`, `"solar_profile"`,
`Dict[str, ndarray]`	`"outdoor_temp_profile"` — all `jnp.float32` arrays of shape
`Dict[str, ndarray]`	`(episode_len,)`.

Notes

Real data (cpu_profile): loaded from the DC workload parquet file when available.

Real data (solar_profile): loaded from the GB generation-by-type solar trace when available.

Real data (price_profile): loaded from the GB MID market trace when available.

Deterministic adapter (outdoor_temp_profile): synthetic sine curve — no weather manifest exists yet.

`powerzoojax.data.dc_microgrid_profiles.apply_ood_transform(params, scenario, *, drought_factor=0.2, temp_delta=5.0, dg_derating_factor=0.6, sla_slack=1.2, data_dir=None, manifest_dir=None)` ¶

Apply an OOD scenario transform, returning a modified DCMicrogridParams.

Parameters:

Name	Type	Description	Default
`params`	`'DCMicrogridParams'`	Base params — should have non-None profiles for transforms that modify them. If profiles are None, a synthetic fallback is generated automatically.	required
`scenario`	`str`	One of `VALID_OOD_SCENARIOS`.	required
`drought_factor`	`float`	Solar scaling for `renewable_drought` (default 0.2).	`0.2`
`temp_delta`	`float`	Temperature offset [°C] for `cooling_stress` (default 5.0).	`5.0`
`dg_derating_factor`	`float`	DG capacity multiplier for `dg_derating` (0.6).	`0.6`
`sla_slack`	`float`	New deadline slack for `sla_tighten` (default 1.2).	`1.2`
`data_dir`	`Optional[str]`	Override data dir for workload_swap / workload_shock.	`None`
`manifest_dir`	`Optional[str]`	Override manifest dir for workload_swap / workload_shock.	`None`

Returns:

Type	Description
`'DCMicrogridParams'`	New `DCMicrogridParams` with the transform applied.

Raises:

Type	Description
`ValueError`	Unknown scenario.

Data¶

Semantic loading¶

Stable signal names¶

Loader, manifest, registry¶

powerzoojax.data.data_loader.DataLoader(data_dir=None, manifest_dir=None) ¶

registry property ¶

load_jax_profiles(signals, *, source=None, region=None, start_date=None, end_date=None, resample=None, time_alignment=None, interpolation='linear', dtype=jnp.float32) ¶

load_jax_array(df, columns=None, dtype=jnp.float32) ¶

load_signals(signals, *, source=None, region=None, start_date=None, end_date=None, resample=None, time_alignment=None, interpolation='linear') ¶

load_actual_series(signals, **kwargs) ¶

load_forecast_panel(signals, *, source=None, region=None, start_date=None, end_date=None) ¶

list_available_datasets() ¶

get_metadata(dataset_name) ¶

get_available_columns(dataset_name) ¶

get_date_range(dataset_name) ¶

find_datasets_for_columns(columns) ¶

load_data(dataset_name=None, columns=None, start_date=None, end_date=None, validate_dates=True, merge_how='inner', resample=None, interpolation='linear') ¶

print_dataset_info(dataset_name) ¶

get_summary() ¶

signals property ¶

raw_columns_needed property ¶

from_json(path) classmethod ¶

to_dict() ¶

powerzoojax.data.registry.DatasetRegistry(manifest_dir=None) ¶

find_by_signal(signal, source=None, data_type=None) ¶

resolve_signals(signals, source=None) ¶

powerzoojax.data.alignment.TimeAligner ¶

align_calendar(df, sim_start, sim_end, align_from=None, time_col='datetime') staticmethod ¶

align_profile(df, sim_start, sim_end, resolution, time_col='datetime') staticmethod ¶

tile_profile_jax(values, n_needed) staticmethod ¶

Parameters¶

Returns¶

align(df, *, time_mode, sim_start, sim_end, align_from=None, resolution='30min', time_col='datetime') classmethod ¶

Frozen splits¶

powerzoojax.data.splits.gb_windows() ¶

powerzoojax.data.splits.ausgrid_windows() ¶

Ausgrid utilities¶

powerzoojax.data.ausgrid_utils.get_ausgrid_split(role) ¶

powerzoojax.data.ausgrid_utils.get_feeder_substations(feeder, role='train') ¶

powerzoojax.data.ausgrid_utils.select_full_coverage_substations(df, start, end, *, max_gap_steps=2, substation_col='region', time_col='datetime', resolution='15min') ¶

Non-stationary sampling¶

powerzoojax.data.nonstationary.EpisodeConfig ¶

powerzoojax.data.nonstationary.NonstationarySampler(total_steps, total_episodes=500, *, max_steps=48, drift_rate=0.0003, rolling_sigma_steps=720.0) ¶

sample(episode_idx, key, *, mode='train') ¶

sample_batch(start_idx, batch_size, key, *, mode='train') ¶

powerzoojax.data.nonstationary.apply_drift(load_profiles_p, load_profiles_q, drift_factor) ¶

DC microgrid profiles¶

powerzoojax.data.dc_microgrid_profiles.cycle_profile(arr, episode_len, start_step=0) ¶

powerzoojax.data.dc_microgrid_profiles.make_all_synthetic_profiles(episode_len=DC_EPISODE_LEN, steps_per_day=DC_STEPS_PER_DAY) ¶

powerzoojax.data.dc_microgrid_profiles.load_workload_profiles(source='google', episode_len=DC_EPISODE_LEN, start_step=0, data_dir=None, manifest_dir=None, strict=True, require_real_data=False) ¶

powerzoojax.data.dc_microgrid_profiles.apply_ood_transform(params, scenario, *, drought_factor=0.2, temp_delta=5.0, dg_derating_factor=0.6, sla_slack=1.2, data_dir=None, manifest_dir=None) ¶

`powerzoojax.data.data_loader.DataLoader(data_dir=None, manifest_dir=None)` ¶

`registry` `property` ¶

`load_jax_profiles(signals, *, source=None, region=None, start_date=None, end_date=None, resample=None, time_alignment=None, interpolation='linear', dtype=jnp.float32)` ¶

`load_jax_array(df, columns=None, dtype=jnp.float32)` ¶

`load_signals(signals, *, source=None, region=None, start_date=None, end_date=None, resample=None, time_alignment=None, interpolation='linear')` ¶

`load_actual_series(signals, **kwargs)` ¶

`load_forecast_panel(signals, *, source=None, region=None, start_date=None, end_date=None)` ¶

`list_available_datasets()` ¶

`get_metadata(dataset_name)` ¶

`get_available_columns(dataset_name)` ¶

`get_date_range(dataset_name)` ¶

`find_datasets_for_columns(columns)` ¶

`load_data(dataset_name=None, columns=None, start_date=None, end_date=None, validate_dates=True, merge_how='inner', resample=None, interpolation='linear')` ¶

`print_dataset_info(dataset_name)` ¶

`get_summary()` ¶

`signals` `property` ¶

`raw_columns_needed` `property` ¶

`from_json(path)` `classmethod` ¶

`to_dict()` ¶

`powerzoojax.data.registry.DatasetRegistry(manifest_dir=None)` ¶

`find_by_signal(signal, source=None, data_type=None)` ¶

`resolve_signals(signals, source=None)` ¶

`powerzoojax.data.alignment.TimeAligner` ¶

`align_calendar(df, sim_start, sim_end, align_from=None, time_col='datetime')` `staticmethod` ¶

`align_profile(df, sim_start, sim_end, resolution, time_col='datetime')` `staticmethod` ¶

`tile_profile_jax(values, n_needed)` `staticmethod` ¶

`align(df, *, time_mode, sim_start, sim_end, align_from=None, resolution='30min', time_col='datetime')` `classmethod` ¶

`powerzoojax.data.splits.gb_windows()` ¶

`powerzoojax.data.splits.ausgrid_windows()` ¶

`powerzoojax.data.ausgrid_utils.get_ausgrid_split(role)` ¶

`powerzoojax.data.ausgrid_utils.get_feeder_substations(feeder, role='train')` ¶

`powerzoojax.data.ausgrid_utils.select_full_coverage_substations(df, start, end, *, max_gap_steps=2, substation_col='region', time_col='datetime', resolution='15min')` ¶

`powerzoojax.data.nonstationary.EpisodeConfig` ¶

`powerzoojax.data.nonstationary.NonstationarySampler(total_steps, total_episodes=500, *, max_steps=48, drift_rate=0.0003, rolling_sigma_steps=720.0)` ¶

`sample(episode_idx, key, *, mode='train')` ¶

`sample_batch(start_idx, batch_size, key, *, mode='train')` ¶

`powerzoojax.data.nonstationary.apply_drift(load_profiles_p, load_profiles_q, drift_factor)` ¶

`powerzoojax.data.dc_microgrid_profiles.cycle_profile(arr, episode_len, start_step=0)` ¶

`powerzoojax.data.dc_microgrid_profiles.make_all_synthetic_profiles(episode_len=DC_EPISODE_LEN, steps_per_day=DC_STEPS_PER_DAY)` ¶

`powerzoojax.data.dc_microgrid_profiles.load_workload_profiles(source='google', episode_len=DC_EPISODE_LEN, start_step=0, data_dir=None, manifest_dir=None, strict=True, require_real_data=False)` ¶

`powerzoojax.data.dc_microgrid_profiles.apply_ood_transform(params, scenario, *, drought_factor=0.2, temp_delta=5.0, dg_derating_factor=0.6, sla_slack=1.2, data_dir=None, manifest_dir=None)` ¶