Error in predict_insample() with Variable-Length Timeseries

[aitirga]

What happened + What you expected to happen

When using the predict_insample() method on an NHITS model with a DataFrame containing multiple timeseries of varying lengths, the following error occurs

Traceback (most recent call last):
...
/site-packages/neuralforecast/core.py", line 1410, in predict_insample
    fcsts[:, col_idx : (col_idx + output_length)] = model_fcsts
    ~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ValueError: could not broadcast input array from shape (960,1) into shape (660,1)

Versions / Dependencies

I am using a macos OS with python 3.11.10 and neuralforecast 1.7.6

Reproduction script

import neuralforecast as nf
from neuralforecast.models import NHITS
import pandas as pd

if __name__ == '__main__':
    df = pd.DataFrame()
    df['ds'] = pd.date_range(start='1/1/2018', periods=100, freq='D')
    df['y'] = range(100)
    df['unique_id'] = 'station_1'
    df_additional = pd.DataFrame({
        'ds': pd.date_range(start='1/1/2017', periods=40, freq='D'),
        'y': range(40),
        'unique_id': ['station_2']*40,
    })
    df = pd.concat([df, df_additional], ignore_index=True)
    model = NHITS(h=5, input_size=7, accelerator='cpu', max_steps=5)
    nf_model = nf.NeuralForecast(models=[model], freq='D')
    nf_model.fit(df)
    nf_model.predict_insample()

Issue Severity

High: It blocks me from completing my task.

[marcopeix]

Can reproduce the error, thanks for pointing it out! We'll work on a fix.

[jmoralez]

We already have some issues about this:

• Intermittent data shortens predict_insample() date range #718
• 使用iTransformer遇到问题 #1000
• 样本内预测predict_insample无法使用 #1056

It'd be good to choose one and concentrate the discussion there