https://github.com/nixtla logo
#neural-forecast
Title
# neural-forecast
a

Afiq Johari

07/11/2023, 3:58 AM
Hi Nixtla team, Is there any updated pretrained models for the following? • Pretrained N-HiTS M4 Monthly From https://nixtla.github.io/neuralforecast/examples/save_load_models.html, I understand there should be two files • `[model_name]_[suffix].ckpt`: Pytorch Lightning checkpoint file with the model parameters and hyperparameters. • `[model_name]_[suffix].pkl`: Dictionary with configuration attributes. From Pretrained N-HiTS M4 Monthly, it seems we can no longer find
[model_name]_[suffix].pkl
I'd like to use the pretrained model since I don't have much access to the GPU. Currently trying to use google colab instead in the meanwhile, but probably will face some limitations.
m

Max (Nixtla)

07/11/2023, 6:47 AM
Hi Afiq! Thanks for the interest. We are slowly making some new and bigger pre-trained models available to some users. The idea is to have something like an api for time series available. We would love to understand your use case better.
f

fede (nixtla) (they/them)

07/11/2023, 6:48 AM
Hey @Afiq Johari! The pretrained models in the repo only work with neuralforecast<1.0.0. Please follow this notebook (https://github.com/Nixtla/transfer-learning-time-series/blob/main/nbs/Transfer_Learning.ipynb) and use “pip install “neuralforecast<1.0.0”“. 🙂
🙏 1
a

Afiq Johari

07/11/2023, 6:52 AM
Hi @Max (Nixtla), thank you for your response. I would like to clarify that I am interested in utilizing the pre-trained model (with a monthly frequency) for short-term forecasting or nowcasting, specifically for a 5-month horizon. Currently, I am unable to use the API due to data privacy concerns. Therefore, I would like to load the pre-trained models directly into my server environment.
@fede (nixtla) (they/them) thanks for this tips!
3 Views