Predictor

The Predictor class provides the inference interface for trained models.

Class Definition

class Predictor:
    """
    Inference interface for trained RESOLVE models.

    Loads saved checkpoints and produces predictions on new data.
    """

Constructor

Predictor(model, species_encoder, scalers, device="auto")

Create a predictor from model components.

Parameters:

Parameter	Type	Description
model	ResolveModel	Trained model
species_encoder	SpeciesEncoder	Fitted species encoder
scalers	dict	Fitted scalers
device	str	Device for inference

Class Methods

load(path, device="auto")

Load predictor from saved checkpoint.

Parameters:

Parameter	Type	Default	Description
path	str \| Path	required	Path to checkpoint file
device	str	"auto"	Device for inference

Returns: Predictor

Example:

predictor = Predictor.load("model.pt", device="cpu")

Methods

predict(dataset, return_latent=False, output_space="raw")

Generate predictions for a dataset.

Parameters:

Parameter	Type	Default	Description
dataset	ResolveDataset	required	Dataset to predict on
return_latent	bool	False	Include latent representations
output_space	str	"raw"	Output space for regression ("raw" or "transformed")

Returns: ResolvePredictions

Example:

predictions = predictor.predict(dataset, return_latent=True)

get_embeddings(dataset)

Get latent embeddings for all plots.

Returns: np.ndarray of shape (n_plots, latent_dim)

embeddings = predictor.get_embeddings(dataset)

get_genus_embeddings()

Get learned genus embedding weights.

Returns: np.ndarray of shape (n_genera, genus_emb_dim)

genus_emb = predictor.get_genus_embeddings()

get_family_embeddings()

Get learned family embedding weights.

Returns: np.ndarray of shape (n_families, family_emb_dim)

family_emb = predictor.get_family_embeddings()

---

ResolvePredictions

Container for model predictions.

Attributes

Attribute	Type	Description
predictions	dict[str, np.ndarray]	Predictions per target
plot_ids	np.ndarray	Plot identifiers
latent	np.ndarray \| None	Latent representations (if requested)

Methods

__getitem__(target)

Get predictions for a specific target.

area_preds = predictions["area"]

to_polars()

Convert predictions to Polars DataFrame.

Returns: pl.DataFrame

df = predictions.to_polars()

to_csv(path)

Save predictions to CSV file.

predictions.to_csv("predictions.csv")