Skip to content

ENH : be able to use temporal or non temporel features in DL workflow.

Arthur Vincent requested to merge issue#618_pytorch_external_features into develop

MR to control output user defined functions.

TU/TI : pass Documentation about deeplearning and external features has been updated.

output must be as :

from typing import Union
import numpy as np
from iota2.learning.utils import I2Label, I2TemporalLabel
def my_function(self) -> tuple[np.ndarray, Union[I2Label, I2TemporalLabel, None]]:
    ...
  • if output array has the wrong dimension, an exception will be raised.
  • if labels does not respect Union[None, Union[I2Label, I2TemporalLabel, None]]], an exception will be raised.
  • default values for labels are 'custfeat_numXbN' where X is the user function's index and N the feature index.

We can no longer pre-instanciate the neural network at iota2 launch.

Edited by Arthur Vincent

Merge request reports