Mismatch between output_size and inputs of predict method translates in missing values in the prediction method

kdgutier / esrnn_torch

MIT License

156 stars 44 forks source link

Mismatch between output_size and inputs of predict method translates in missing values in the prediction method #21

Closed CoteDave closed 3 years ago

CoteDave commented 3 years ago

When using multiple timeseries ID, there seem to be missing values after the predict:

CoteDave commented 3 years ago

Resolved!

kdgutier commented 3 years ago

Hi @CoteDave, something similar happened before to a user, the reason behind it was a problem with the date stamps generated in pandas for the prediction merge into the final data frame. Another time it was a mismatch between expected output_sizes of the test and the output_size of the model. What was the problem for you?

CoteDave commented 3 years ago

Sure! I was thinking that the model automaticly fills all the dates in the data given to the predict method. Unfortunatly, I discovered the "output_size" hyper-parameter that was set to 3... so even if my predict data has 12 dates per timeserie in it, it only returned 3 predictions per timeserie. I changed it to 12 and everything was fine. My bad!