something bad happened to cross-validation with 1.7.0? getting errors KeyError: ‘unique_id’ while with 1.6.4 same code works well

Hey. Can you provide the full stacktrace?

Traceback (most recent call last):

File "/usr/local/lib/python3.10/site-packages/pandas/core/indexes/base.py", line 3805, in get_loc

return self._engine.get_loc(casted_key)

File "index.pyx", line 167, in pandas._libs.index.IndexEngine.get_loc

File "index.pyx", line 196, in pandas._libs.index.IndexEngine.get_loc

File "pandas/_libs/hashtable_class_helper.pxi", line 7081, in pandas._libs.hashtable.PyObjectHashTable.get_item

File "pandas/_libs/hashtable_class_helper.pxi", line 7089, in pandas._libs.hashtable.PyObjectHashTable.get_item

KeyError: 'unique_id'

I mean one that includes neuralforecast code

That’s the only one i’m getting running the same code

Are you providing the id as the index in df?

yes

Can you set it as a column?

Wait, you mean do i have personal index? Then - no.

what are the names of your columns?

df = pd.read_csv(temp_dataset_path)

df = df.reset_index(drop=True)

df['ds'] = <http://pd.to|pd.to>_datetime(df['ds'])

If you can provide a small reproducible example I can help further, otherwise it's very hard to guess what might be going on. We test several times the cross validation in our CI

Oh, there is not much to provide tbh, literally very basic and simple usage where i do have file csv file with three columns. Loading file via pandas and triggering nf.cross_validation

The following works fine for me:

from neuralforecast import NeuralForecast
from neuralforecast.models import NHITS
from neuralforecast.utils import AirPassengersDF

nf = NeuralForecast(models=[NHITS(h=12, input_size=24, max_steps=1)], freq='M')
nf.cross_validation(df=AirPassengersDF, n_windows=3, step_size=1)

That's a small example. If you can provide one like that where there's an error I can help further

The only difference is that i’m using Auto models

If it's pretty much the same can you please just create the example with your modifications? I don't know which arguments you're using

nf = NeuralForecast(

models=[

AutoMLP(h=nf_horizon, config=None, loss=HuberMQLoss(), num_samples=1, backend='optuna', verbose=False),

AutoNBEATS(h=nf_horizon, config=None, loss=HuberMQLoss(), num_samples=1, backend='optuna', verbose=False),

],

freq = nf_freq,

local_scaler_type = "robust"

)

Your example runs fine for me as well, is the error actually coming from neuralforecast or further down the line in your code?

Error comes when nf.cross_validation is executed

Yeah that doesn't help. I ran this:

from neuralforecast import NeuralForecast
from neuralforecast.auto import AutoMLP, AutoNBEATS
from neuralforecast.utils import AirPassengersDF

def config(trial):
    return {
        'input_size': 24,
        'max_steps': 1,
    }

nf = NeuralForecast(
    models=[
        AutoNBEATS(h=12, config=config, backend='optuna', num_samples=1),
        AutoMLP(h=12, config=config, backend='optuna', num_samples=1)
    ],
    freq='M',
)
print(nf.cross_validation(df=AirPassengersDF, n_windows=3, step_size=1))

without any issues. If you can't provide an example like that where there's an error or a full stacktrace (including neuralforecast code) I can't help any further

Ugh, i understand… Really something strange. I really appreciate your time and help. Will go to debug more and let you know. Thank you very much!

There's an important change in 1.7.0 with respect to the index (we're deprecating it from the input and output) so if you set

os.environ['NIXTLA_ID_AS_COL'] = '1'

you'll get the id as a column instead of the index in the predict and cross_validation outputs, which might help

Oooooh waaait, where it is written???

You should get future warnings if you're not setting it

It seems there is some weird stuf fhere

If I change my example to

df=AirPassengersDF.set_index('unique_id')

I get the key error. So you should be able to fix it by doing

df.reset_index()

. We'll fix it on our side as well but in the meantime that should work for you

i am doing that :))))))))

Maybe you saved the df with the index and when you drop it it goes away, so maybe just

df = df.reset_index()

?

Will try rn

I guess you have to reset the index of cv_df as well

Maaaan, we are gonna break nf documentation here together :)))

yeah sorry for the troubles. We have this issue to update that tutorial and we'll also fix the CV for the id as index case

Don’t call it troubles. NF is literally the holigrail of forecasting.

Wow, just came over here for this. Was getting the same KeyError: ‘unique_id’. So reset the index right?

Yes, just make sure the id is a column. We'll fix this in the next release, which should be next week at the latest