Combine multiple GeoJSON files into a single dataset. IMPORTANT: This function respects existing labels in the input files. It does NOT

Load model from JSON file.

Add or update label for a detection tile_id.

Set up mock layer manager widget attributes for testing.

Report progress if callback is provided.

Load and split training data. Parameters ---------- test_fraction : float Fraction of data to use for te

Create fold assignments based on spatial clustering of positives. Algorithm: 1. Buffer positive point geometries and union into contiguo

Run batch inference over all embeddings. Parameters ---------- probability_threshold : float Minimum pro

Randomly sample negative examples from DuckDB, excluding positives. This provides a simpler alternative to LULC-based stratified sampling.

Train the classifier. Parameters ---------- X_train : np.ndarray Training features, shape (n_samples, n_

Perform k-fold cross-validation with pre-assigned folds. Parameters ---------- X : np.ndarray Feature matrix (n_samples, n_f

Parse MGRS tile_id to get UTM CRS.

Add an XYZ tile layer overlay.

Enter detection review mode with provided GeoJSON data.

Returns 's3' or 'gs' if the path is a cloud path, otherwise None.

Check if path is a Google Cloud Storage path. Args: path: Path to check Returns: True if path is GCS URL, Fa

Return human-readable timing summary.

Toggle label assignment for a point and return resulting state.

Install and load the requested DuckDB extension.

Predict probability of positive class. Parameters ---------- X : np.ndarray Features, shape (n_samples,

Reset all mutable state.

Save model to JSON format.

Extract tile IDs from the results grid children in display order.

Refresh detection layer with current threshold and labels.

Score a batch of embeddings. Parameters ---------- embeddings : np.ndarray Batch of embeddings, shape (n

Find nearest tile_id for each point using spatial matching.

Evaluate the classifier on test data. Parameters ---------- X_test : np.ndarray Test features, shape (n_

Convert geographic coordinates to pixel coordinates. Args: lon: Longitude lat: Latitude geo_bounds: (min_lon, min_lat, ma

Return all MGRSTileIds whose footprints intersect the *search_geometry*. Parameters ---------- search_geometry : shapely.geometry

Get total number of embeddings in database.

Remove an overlay layer by name.

Update the active label mode.

Set the opacity of an overlay layer.

Sample random rows excluding positive tile_ids.

Fetch metadata for detected IDs. Parameters ---------- detections : List[Tuple[int, float]] List of (id,

Get cloud-masked Sentinel-2 collection using CloudScore+ for quality filtering. Args: aoi: Earth Engine geometry defining area of interes

Load all training data without split, with geometries for spatial CV. Parameters ---------- random_state : int

Return HTML for the status bar.

Sample random negatives using positives from a GeoJSON file. This is a convenience function that extracts tile_ids from a GeoJSON file a

Union overlapping tiles into larger polygons. Parameters ---------- tiles_gdf : gpd.GeoDataFrame GeoData

Detect embedding dimension from first row.

Fetch embeddings from DuckDB for given tile_ids.

Generator yielding (ids, embeddings) batches using fetchdf(). Yields ------ ids : np.ndarray Array of em