Closed AzulGarza closed 2 years ago
Since n_layers
and n_blocks
are parameters of the model, I think the best solution is to rename the hyperopt space parameters to constant_n_layers
and constant_n_blocks
,
The same applies for n_mlp_units
.
After getting the best hyperopt trial using
nf.experiments.utils.hyperopt_tunning
, I wanted to use the mc provided in the trial results innf.experiments.utils.instantiate_nbeats
. I got the following error:The problem arises from these lines:
https://github.com/Nixtla/neuralforecast/blob/09ce6d0075bbf287d78483d04f107dec7505fc5c/neuralforecast/experiments/utils.py#L270-L273
Initially,
n_blocks
andn_layers
are integers (denoting the number of blocks and layers). But later,nf.experiments.utils.instantiate_nbeats
converts them to a list of integers. Therefore, themc
dictionary becomes unusable for reuse bynf.experiments.utils.instantiate_nbeats
.