[N-HiTS Experiment]: code for the other datasets

[AlexMV12]

Description

The experiments/long_horizon folder contains the code to reproduce the N-HiTS experiment reported in the original paper only for the datasets ETT. I tried to slightly change the code to adapt the code to another dataset (i.e., ILI), but it does not work (and I am unsure on how to fix this, given that also two files for loading the datasets exist in the repo).

It would be a great value being able to reproduce the experiment of the N-HiTS paper on all the datasets presented in the paper. Thanks for the nice work!

Link

No response

[cchallu]

Hi @AlexMV12 . Can you provide your code or the error, please?

[AlexMV12]

Hello @cchallu, absolutely.

Actually, after a bit of work, I managed to make it work with all the other datasets present in the paper. I will give you the code, so that you can choose to integrate it in the repo and to make it available to everyone looking into reproducing NHiTS's results.

Note that the following code works:

1. only on the main branch of this repo;
2. only on the main branch of datasetsforecast, because it need the LongHorizonInfo class instead of the LongHorizonInfo2 class.

With this in mind, I modified the expriments/long_horizon/run_nhits.py script as follow (the first 60 rows):

import os
os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1"

import argparse
import pandas as pd

from ray import tune

from neuralforecast.auto import AutoNHITS
from neuralforecast.core import NeuralForecast

from neuralforecast.losses.pytorch import MAE, HuberLoss
from neuralforecast.losses.numpy import mae, mse
from datasetsforecast.long_horizon import LongHorizon, LongHorizonInfo
from datasetsforecast.long_horizon2 import LongHorizon2, LongHorizon2Info

import logging
logging.getLogger("pytorch_lightning").setLevel(logging.WARNING)

if __name__ == '__main__':

    # Parse execution parameters
    verbose = True
    parser = argparse.ArgumentParser()
    parser.add_argument("-horizon", "--horizon", type=int)
    parser.add_argument("-dataset", "--dataset", type=str)
    parser.add_argument("-num_samples", "--num_samples", default=5, type=int)

    args = parser.parse_args()
    horizon = args.horizon
    dataset = args.dataset
    num_samples = args.num_samples

    # Load dataset
    if dataset in ['ETTm1', 'ETTm2', 'ETTh1', 'ETTh2']:
        Y_df = LongHorizon2.load(directory='./data/', group=dataset)
    else:
        Y_df, _, _ = LongHorizon.load(directory='./data/', group=dataset)
        Y_df['ds'] = pd.to_datetime(Y_df['ds'])

    if dataset != 'ILI':
        assert horizon in [96, 192, 336, 720]
    else:
        assert horizon in [24, 36, 48, 60]

    freq = LongHorizonInfo[dataset].freq
    n_time = len(Y_df.ds.unique())

    if dataset in ['ETTm1', 'ETTm2', 'ETTh1', 'ETTh2']:
        val_size = LongHorizon2Info[dataset].val_size
        test_size = LongHorizon2Info[dataset].test_size
    else:
        val_size = int(.2 * n_time)
        test_size = int(.2 * n_time)

...

This works on the ILI dataset and the result looks good. I will try to run the rest of the dataset, but now everything should work.

[cchallu]

Awesome, thanks for sharing the code!