app.models module#

Declarative SQLAlchemy ORM models used across the application.

This module defines the core database entities for the platform: coordinates for data collection, time-stamped weather observations, a singleton table for training status, and an append-only log of ML training runs. These models are the canonical schema definition for the application and are used by both the FastAPI backend and the standalone Slurm-executed training jobs for reading and writing domain data.

Notes#

Primary role: provide ORM mappings for coordinates, observations, training status, and training logs bound to app.database.Base.
Key dependencies: a configured SQLAlchemy engine via app.database.engine and corresponding session factories. The schema is created via app.database.ensure_database_schema().
Invariants: WeatherObservation uses a composite primary key (timestamp, latitude, longitude). TrainingStatus is treated as a singleton with primary key id=1. TrainingLog.horizon is a non-null string key identifying coordinate+horizon groupings.

Examples#

>>> # Basic query pattern (requires an initialized DB)
>>> from app.database import SessionLocal, ensure_database_schema
>>> from app.models import Coordinate, TrainingLog
>>> ensure_database_schema()
>>> with SessionLocal() as session:
...     count = session.query(Coordinate).count()
...     latest = (session.query(TrainingLog)
...               .order_by(TrainingLog.timestamp.desc())
...               .first())

class app.models.Coordinate(*args: Any, **kwargs: Any)[source]#

Bases: Base

Geographic point used for weather data collection.

Attributes:

idint: Surrogate primary key.
latitudefloat: Coordinate latitude in decimal degrees.
longitudefloat: Coordinate longitude in decimal degrees.
labelstr | None: Optional human-readable label for the coordinate.
is_centralbool: Marks the central coordinate used as a reference point.

Notes

No uniqueness constraint is enforced at the ORM level for (latitude, longitude); duplicates are possible unless prevented by a database constraint or application logic.

class app.models.TrainingLog(*args: Any, **kwargs: Any)[source]#

Bases: Base

Append-only log of ML training runs and scores.

Each row represents one training execution for a given horizon and (optionally) a specific coordinate. Scores from both a Scikit-learn model and a PyTorch model are recorded, along with the number of data points used.

Attributes:

idstr: Primary key string. The training pipeline typically assigns a uuid4 string explicitly when inserting a row. The default defined here is a string constant created at import time and should be overridden by callers to avoid collisions.
timestampdatetime: Completion time of the training run in UTC.
horizonstr: Non-empty key identifying the grouping (often "<coord>_<horizon_label>").
sklearn_scorefloat: R^2 score from the Scikit-learn model.
pytorch_scorefloat: R^2 score from the PyTorch model.
data_countint: Number of samples used for training/evaluation for this run.
coord_latitudefloat | None: Coordinate latitude associated with the run, if available.
coord_longitudefloat | None: Coordinate longitude associated with the run, if available.
horizon_labelstr | None: Human-friendly label for the horizon (e.g., "5min", "1h").

class app.models.TrainingStatus(*args: Any, **kwargs: Any)[source]#

Bases: Base

Snapshot of the current ML training state.

This singleton table reflects whether a training job is running, when the last successful training completed, and which horizon is currently being processed. The application typically ensures that there is exactly one row with id=1 and updates it transactionally during training flows.

Attributes:

idint: Primary key. Conventionally set to 1 to model a singleton row.
is_trainingbool: Flag indicating whether a training job is currently running.
last_trained_atdatetime | None: Timestamp of the last completed training job in UTC, if any.
train_countint: Monotonic counter of completed training runs.
current_horizonstr | None: Human-readable horizon label (e.g., "5min") or status message.

class app.models.WeatherObservation(*args: Any, **kwargs: Any)[source]#

Bases: Base

Nowcast weather values for a specific time and location.

This table mirrors the structure of MET Norway Nowcast 2.0 fields consumed by the application. The composite primary key (timestamp, latitude, longitude) uniquely identifies each observation in time and space.

Attributes:

timestampdatetime: Observation timestamp in UTC (part of the composite primary key).
latitudefloat: Coordinate latitude in decimal degrees (part of the primary key).
longitudefloat: Coordinate longitude in decimal degrees (part of the primary key).
air_temperaturefloat | None: Air temperature in degrees Celsius.
wind_speedfloat | None: Wind speed in meters per second.
wind_directionfloat | None: Wind direction in degrees (from which the wind is coming).
cloud_area_fractionfloat | None: Fraction of the sky covered by clouds (0–1).
precipitation_amountfloat | None: Precipitation amount in millimeters for the interval.
is_imputedbool: Whether this record was imputed by preprocessing routines.

app.models module#

See Also#

Notes#

Examples#