Adding feature/s for model training in GluonTs - DeepAr

[Msaleh87]

I know that this might seems like a similar question, I read almost read all of the questions about this issue, and they look similar, but for some reason, I can't get it right. I tried to follow the documentation of the extended forecasting tutorial, but it's not clear at all.
I created a data frame that has 3 columns:
1- electrical load historical data with 5 min time step resolution
2- timestamp
3- weekday or weekend (this is the feature I want to add)

Here is my code:

estimator = DeepAREstimator(freq = '5min',
                           context_length = 7 * 24 * 12,  
                           prediction_length = 1*24*12,
                           num_layers = 2,
                           num_cells = 64,
                           cell_type = 'lstm',
                           use_feat_dynamic_real=True,
                           trainer = Trainer(epochs= 10, num_batches_per_epoch=64))

from gluonts.dataset.common import ListDataset
training_data = ListDataset([{"start": df_input.index[0],
                              "target": df_input['Load-kW'][:(12*30*24*12)], # this is just the range of my target
                              "dynamic_feat": df_input['DofW'][:(12*30*24*12)]}],
                           freq = "5min")


predictor = estimator.train(training_data = training_data)

After the last command I get the following error: GluonTSDataError: Array 'dynamic_feat' has bad shape - expected 2 dimensions, got 1.

I am just trying to add one feature. Also, when I add a dynamic feature, do I need to make it's length = (target length + forecast length) or just the target length?

So I fixed some of the initial mistakes, which is

training_data = ListDataset([{"start": df_input.index[0],
                              "target": df_input['Load-kW'][:'2022-03-07 04:50:00'],
                              "feat_dynamic_real": Feat[:'2022-03-07 04:50:00']}],
                           freq = "5min")

Feat here is a time-stamped data frame with the feature only
Now I am getting this new error: ValueError: all the input array dimensions except for the concatenation axis must match exactly.
Pleas help!

[Msaleh87]

My bad it should have been "feat_dynamic_real":[ Feat[:'2022-03-07 04:50:00']]. Was missing square brackets.