learn.property_prediction.init_loss_fn#
- init_loss_fn(error_fn)[source]#
Returns a loss function to optimize model parameters.
Signature of error_fn:
error = error_fn(predictions, batch, mask)
where mask has the same shape as species to mask padded particles.
- Parameters:
model – Molecular property prediction model (Haiku apply_fn).
error_fn – Error model quantifying the discrepancy between predictions and respective targets.