https://github.com/nixtla logo
Join Slack
Powered by
Hi <@U03DDFKMT0X> , if you are using the auto wrap...
# neural-forecast
k
Hi @Chris Gervais , if you are using the auto wrappers of the experiment utils: You can parse the predictions from the trials dictionary, see the end part of this N-HiTS Colab example
c
i think we're able to unpack the results no problem, but we're having issues assigning timestamps to the predictions
k
sorry for the inconvenience we are working on this pull request to assign the predictions time stamps, we hope to have it ready for Thursday https://github.com/Nixtla/neuralforecast/pull/237
If you are working on a balanced dataset @Chris Gervais you could reshape the numpy vectors and assign them a pandas datestamps manually which is what we have done in the past
c
no worries! i think we're having issues backing out the timestamps because the val + test masks are all 1's and they're only valid for their respective data splits, whereas the meta data that comes back (containing timestamps) is for the entire train / val / test dataset
k
We will ping you on Thursday, for the updated predict functions with the datestamps This has been bugging us too
c
👍 thanks!
do you plan on releasing a new tag shortly after that MR goes in or do you think it'll be a little while before 
v0.1.1
goes out?
k
We plan to integrate the predictions with dates into main by Thursday
👍 2
Hey @Chris Gervais We uploaded the working 
NHITS.predict
method Can you take a look? 1) Installation: pip install git+https://github.com/Nixtla/neuralforecast.git@nbeats_predict 2) NHiTS.predict example: https://github.com/Nixtla/neuralforecast/blob/nbeats_predict/nbs/models_nhits__nhits.ipynb 3) Predict branch: https://github.com/Nixtla/neuralforecast/blob/nbeats_predict
c
do these methods work with 
experiments/utils.py::predict(...)
?
k
We have not integrated the in-sample 
NHITS.predict
function in the 
experiments/utils.py
The method is creating the predictions, parsing them and returning Y_hat_df in long format.
c
I noticed the MR was closed recently, is this feature planned for a future release?
btw we've tested this out on our end and it works great 👍
k
Good to know 👍
c
do you have any guidance on when to use 
.predict
vs 
.forecast
?
seems like they have similar outputs now
k
We intended forecast to be used for outsample predictions
And predict to be used with insample predictions
c
👍 ah okay thanks!
Open in Slack
PreviousNext
Powered by