<#C031EJJMH46|>
Hey everyone I'm using nixtla mlfo...
# generals
sergio lopez
06/10/2024, 3:14 AM#C031EJJMH46
Hey everyone I'm using nixtla mlforecasts. And I'm doing a preprocess of the data to add the variables needed. But for some reason I'm having this error. I kinda understand what's happening but I'm not sure how to fix it cause it's an error inside the preprocess function of nixtla:
j
José Morales
06/10/2024, 4:46 PMHey. Can you provide the code that you're running?
s
sergio lopez
06/10/2024, 6:00 PMSure, it's pretty straightforward
You can assume that the dataframe we are passing to preprocess is of the form:
ds, y, unique_id
ds is a datetime field
y is a numeric field
unique_id is a string field
So I check if I haven't had a problem doing the clearing in the dataframe to fit that form. That's not the case, I'm adding little extra screenshot of the data inside categories[cat]
sergio lopez
06/10/2024, 6:03 PMThere's also extra functions for the features. These functions just returns a categorical vector to see if I'm in a seasonal period. I'm adding one example here:
j
José Morales
06/10/2024, 6:06 PMIs there any date feature function that returns a list?
José Morales
06/10/2024, 6:07 PMWe recently changed this process to be more efficient so we only compute the date features for the unique dates and then kind of upsample it back using a pandas series to index the result. If your function doesn't return a pandas series it could cause that error I think
s
sergio lopez
06/10/2024, 6:07 PMOh yes, I have one doing that:
j
José Morales
06/10/2024, 6:08 PMCan you try just doing
return pd.Series(result)s
sergio lopez
06/10/2024, 6:08 PMOh makes sense, I will change that function to return pandas series instead. Actually that was working on a past version of the library. That error is happening on an updated version.
sergio lopez
06/10/2024, 6:10 PMSure, I will come back when I try it. Thanks in advance José!
🙌 1
sergio lopez
06/11/2024, 8:14 PMI works! That was the problem. I change it to return a pandas series an run without hassle. Thank you very much José!
🙌 1