what's the recommended way to generate a prediction dataframe in

v1.0.0

? it looks like we should always be using the

core.NeuralForecast

wrapper now instead of using the model predict methods directly - is that correct?

Hi <!channel>: We have been working on a new version of NeuralForecast to make usage easier. We would love for you to play around and tell us if you want or need something. New Exciting Features: * Updated Getting Started: https://nixtla.github.io/neuralforecast/examples/getting_started_with_nbeats_and_nhits.html * Unified interface for training and predicting * Classes for automatic hyperparameter optimization for different models: https://nixtla.github.io/neuralforecast/models.html * Multi Quantile Forecast * Out of the Box GPU Support * Unified Datasets/Dataloader class for all models * Support for different hyperopt backends * Pytorch lighting support

Hello All, Can we use the Global models (N-BITS, N-HITS, etc.) for the multiple irregular time series?

hello. how do i perform nbeats and nhits multivariate time series?

Hey everyone, this is Hasnain. I hope everyone is doing well. I am working a regression problem with around 60-70 features and have around 2 Million plus data points and one target. I am searching for a high level wrapper for regression problem where I can do experiments with different architectures such as with different backbones to extract features and heads for regression. If there is any open source library then please let me know. Thanks

btw we’re starting to get some pretty conclusive results re: neuralforecast performance as benchmarked against xgboost, catboost, lightgbm etc. @Valeriy not sure if you’ve had a chance to wrap up your analysis but seems pretty clear at this point the conclusion from https://arxiv.org/abs/2106.03253 is bogus

Too much work recently @Chris Gervais so not much time but will be soon starting forecasting project so will pickup again. What DL model is shining?

I can imagine, especially with all the CP excitement! we’re finding consistent lift from both nbeatsx and nhits implementations.

Greetings! Two questions about the excellent neuralforecast packages. 1. Embeddings for categorical variables appear to have been deprecated. Is this temporary? If not, why was this choice made? 2. Are there plans to introduce probabalistic losses for count targets? (i.e. negative binomial, tweedie distributions for count valued time series).

Happy holidays! Can you help me understand how to use nbeats and nhits? as you can see, i have a time series and both models not recognizing the seasonalities. but the autoarima in Nixtla works well with the same data. here is the code for the two models. anything i need to change? #note: horizon = 12 month, it’s at the beginning of the month like ‘2022-02-01’ models = [NBEATS(input_size=2 * horizon, h=horizon, max_epochs=50), NHITS(input_size=2 * horizon, h=horizon, max_epochs=50)] nforecast = NeuralForecast(models=models, freq=‘MS’)

🦃 🎉

Greetings!, I hope everyone is doing well. I am working on a regression problem and I am looking forward to use Transformers for it but before jumping into the implementation and all stuff, I am curious that did any of you use transformers for regression problem. I have around 90 features (floating points) and one target. I couldn't find any paper on transformers for regression problems so please let me know if any of you used transformers for this purpose.

the Nixtla premium: NHiTS from

neuralforecast

consistently outperforming NHiTS from

pytorch-forecasting

in the wild lol

Hi everyone, I really appreciate your great work with neuroforecast ! I am missing in the documentation information on how to use future, lagged and static exogenous variables. Do you have any example?

I have a question about the default scaling behavior. Are exogenous features scaled automatically based on "scaler_type", or just the targets?

Hi, I'm quite new to nixtla, could someone help me to create a comparison model between TFT and AutoARIMA on my data by cross-validation?

Hi. I am new to the neural-forecast package, so I am sorry if I missed it somewhere. Is there a way to set assumption about seasonality for the AutoNHITS? Would it be “n_harmonic” set in the config object passed?

Hi! What would be the best way to use a NeuralForecast object to test out predictions from AutoTFT, AutoNBEATS, Auto NHiTs? I do see a cross validation module which is great but I do not see any .Forecast method. Should I only use .predict () ?

Hi 🙂. Is there a way to set the number of stacks or is that fixed?

I'm trying to compare our hand-rolled cross-validation methods (which uses

sktime

) with the

neuralforecast

native cross-validation but we're running into a merge issue. The culprit seems to be in the core

NeuralForecast

class here:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[56], line 1
----> 1 nf.cross_validation(df=df, static_df=static_df, val_size=val_size)

File ~/Desktop/rtobots/.venv/lib/python3.9/site-packages/neuralforecast/core.py:386, in NeuralForecast.cross_validation(self, df, static_df, n_windows, step_size, val_size, test_size, sort_df, verbose, **data_kwargs)
    383 fcsts_df = pd.concat([fcsts_df, fcsts], axis=1)
    385 # Add original input df's y to forecasts DataFrame
--> 386 fcsts_df = fcsts_df.merge(df, how="left", on=["unique_id", "ds"])
    387 return fcsts_df

and the actual error appears to be related to index types that don't match:

ValueError: You are trying to merge on datetime64[ns] and object columns. If you wish to proceed you should use pd.concat

Replication: • using the exogenous tutorial here https://nixtla.github.io/neuralforecast/examples/exogenous_variables.html • replacing the

.fit()

code chunk with

.cross_validation(df=df, static_df=static_df, val_size=int(len(df) * .2))

Hi can someone help me understand why my TFT model (AutoTFT) never ends compiling on my data while it works fine with AutoNheats from NHITS ? Please 🙏

Hello, May I know what is "unique_id" and how it is being generated in the datasets? Thank you.

Hi, Can someone help me how to use

neuralforecast

to forecast into the future/unseen/out-sample dates?

Hi guys, thanks for the exciting library. I have a question about exogenous variables with different time internals; Let's say I have two past exogenous variables, one is daily, and another one is monthly, and my target is also daily. Would it be possible to use these two variables as past exogenous variables in the neural models? for example, in the case of the TFT model. If this is possible, I wonder how this is done under the hood. If first needs some data processing to align the time intervals, do you have some ideas how this should be done? Thanks

Hi, can someone help me on saving/viewing the predictions on the historical dataset? Whenever I use

.predict()

, it's forecasting into the future. I also want to look at the historical predictions of the model.

Hi Team, Apart from N-BEATS and related models like N-HiTS and N-BITSX, are there any other models or family of models that can be used as a Global time-series model i.e. train on multiple time series to forecast single time series?

Hello, I am trying to run the tutorial https://colab.research.google.com/github/Nixtla/neuralforecast/blob/main/nbs/examples/IntermittentData.ipynb on my local windows machine. I have installed all the necessary packages on a clean python virtual env using

pip install neuralforecast statsforecast s3fs fastparquet

cmd. I am getting the below error:

from neuralforecast.losses.pytorch import DistributionLoss
ImportError: cannot import name 'DistributionLoss' from 'neuralforecast.losses.pytorch' (E:\MT\code\env\lib\site-packages\neuralforecast\losses\pytorch.py)

Can someone please help. python version is

3.10.9

i noticed in https://nixtla.github.io/neuralforecast/examples/exogenous_variables.html that the

weekday

column doesn't get hot encoded when passing to

futur_exog_list = [..., 'weekday']

- was that intentional? we're trying to integrate tsfresh features but seem to be running into issues when we hot encode - still investigating but thought it might be related to exogenous scaling

Hi all, is there a specific model of Neural Forecast you'd recommend to forecast multiple target variables?

is it possible to use NHITS to forecast future periods if we don't have "live"

y

values?