ctf4science.data_module.get_validation_prediction_timesteps#

ctf4science.data_module.get_validation_prediction_timesteps(dataset_name, pair_id, train_split=0.8)#

Return prediction timesteps for the validation split.

Returns a 1D array of absolute physical times (same units and formula as get_prediction_timesteps()) for the held-out validation window. The source of these timesteps depends on pair_id:

  • pair_ids 2, 4 (reconstruction): prediction timesteps equal the full training matrix timesteps.

  • pair_id 8: timesteps of the second training matrix (the held-out validation matrix).

  • pair_id 9: timesteps of the third training matrix.

  • pair_ids 1, 3, 5, 6, 7: the tail of the single training matrix (rows floor(train_split * N) onward).

Parameters:
dataset_namestr

Name of the dataset to load.

pair_idint

Data pair (1–9; special handling for 2, 4, 8, 9).

train_splitfloat, optional

Fraction of data for training in (0, 1). Default is 0.8. Ignored for pair_ids 2, 4, 8, 9.

Returns:
ndarray

1D array of absolute physical times for the validation prediction window, computed as (start_index + i) * delta_t.

Raises:
ValueError

If train_split is not in (0, 1) or pair_id is not in 1–9.