Models

class cloudcasting.models.AbstractModel(history_steps: int)

An abstract class for validating a generic satellite prediction model

check_predictions(y_hat: Float[ndarray, 'batch channels rollout_steps height width']) → None

Checks the predictions conform to expectations

abstract forward(X: Float[ndarray, 'batch channels time height width']) → Float[ndarray, 'batch channels rollout_steps height width']

Abstract method for the forward pass of the model.

Parameters:

X (BatchInputArray) – Either a batch or a sample of the most recent satellite data. X will be 5 dimensional and has shape [batch, channels, time, height, width] where time = {t_{-n}, …, t_{0}} (all n values needed to predict {t’_{1}, …, t’_{horizon}})

Returns:

The models prediction of the future satellite data of shape [batch, channels, rollout_steps, height, width] where rollout_steps = {t’_{1}, …, t’_{horizon}}.

Return type:

ForecastArray

abstract hyperparameters_dict() → dict[str, Any]

Return a dictionary of the hyperparameters used to train the model

class cloudcasting.models.VariableHorizonModel(rollout_steps: int, history_steps: int)