search

Aminsn / SERT2023

code for the paper, "SERT: A Transfomer Based Model for Spatio-Temporal Sensor Data with Missing Values for Environmental Monitoring"
https://arxiv.org/pdf/2306.03042.pdf
MIT License
6 stars 0 forks source link

`build_strats` inputs mismatch in model_fitting.py #2

Open dbuscombe-usgs opened 2 months ago

dbuscombe-usgs commented 2 months ago

Hi, I'd like to report what I think is a bug, when calling build_strats https://github.com/Aminsn/SERT2023/blob/35b6d1ad455a7cb3fc0a0cc0462cef7d1e9e272e/Simulation_study/model_fitting.py#L177

7 inputs are called, and their names differ from the 6 defined in the function https://github.com/Aminsn/SERT2023/blob/35b6d1ad455a7cb3fc0a0cc0462cef7d1e9e272e/Simulation_study/model_fitting.py#L22

so, I'm trying to reconcile N, he, and V here ...

    strats = build_strats(max_len=fore_max_len,
                          num_var=num_var,
                          d=60,
                          N=6,
                          he=6,
                          V=num_var,
                          dropout=0.1)

with num_repeat and num_head defined here

def build_strats(max_len, num_var, d, num_repeat, num_head, dropout):

thanks! I really appreciate this work - cool paper, too! I'm interested in using these methods for time-series imputations for long-term oceanographic time-series in San Francisco Bay.

Aminsn commented 2 months ago

Hi Daniel, thank you for your comments. I’ll look into the issue, but in the meantime, I recommend using my Python package 'sert', which I developed to simplify the installation and application of the model. Here’s the link to the package:

https://github.com/Aminsn/sert

dbuscombe-usgs commented 2 months ago

Hi Amin, thanks! I managed to install that package and ran the toy examples with no problem. I would be trialing the multivariate forecasting model for time-series imputation

I will be in touch again if I have any more questions. Thanks again for the help