pytorch-forecasting output index

[RW74]

I am trying to use the library but I have an issue on how to handle the output from a model prediction, the index shows time_idx starting from 36 and repeating the last values at the last time_idx (encoding_length 72 and prediction_length is 1). If I change the enconding_length half of it goes as start point and the other with repeating values at the end. I can share the full code but it is few thousands lines mostly to handle a yaml interface to define the features targets etc, I can share logs otherwise, so tell me what you think is best. I am not familiar with the library so it could be something in the tsds that is set not properly (I can share the output of these). but the prog is running a training, saving the model and the tsds then running a prediction from the saved model and tsds. what I saw i that after reloading the tsds it is not exactly as the original, it shows this which it was not set for the training:

'categorical_encoders': {'__group_id__group_id': NaNLabelEncoder(add_nan=False, warn=True), 'group_id': NaNLabelEncoder(add_nan=False, warn=True)}, 

this is the output from the index after the predict (encoding length 8, prediction length 1):

------------------------------------------
index_df_from_predict.head():
   time_idx group_id
0         8      SPX
1         9      SPX
2        10      SPX
3        11      SPX
4        12      SPX
5        13      SPX
6        14      SPX
7        15      SPX
8        16      SPX
9        17      SPX
index_df_from_predict.tail():
      time_idx group_id
1004      1012      SPX
1005      1013      SPX
1006      1013      SPX
1007      1013      SPX
1008      1013      SPX
1009      1013      SPX
1010         4      SPX
1011         5      SPX
1012         6      SPX
1013         7      SPX
------------------------------------------
index_df_sorted.head():
      time_idx group_id
1010         4      SPX
1011         5      SPX
1012         6      SPX
1013         7      SPX
0            8      SPX
1            9      SPX
2           10      SPX
3           11      SPX
4           12      SPX
5           13      SPX
index_df_sorted.tail():
      time_idx group_id
1000      1008      SPX
1001      1009      SPX
1002      1010      SPX
1003      1011      SPX
1004      1012      SPX
1005      1013      SPX
1009      1013      SPX
1008      1013      SPX
1007      1013      SPX
1006      1013      SPX

all came out as I am trying to mapping back the datetime to the predicted output.

[fkiraly]

could you add full code with imports? Feel free to use dummy data if you cannot share your actual data (this is preferred in any case, since then we can copy-paste code).

[RW74]

I found the issue :), somehow I had hardcoded min_encoding_length to half of the max encoding. putting both the same the output from index is as expected, starting from 8 and ending without duplicates.