[Bug] ValueError when using scaler.inverse_transform with a three-dimensional array

[Hackerismydream]

Description:
Greetings, and thank you for compiling this insightful survey. As a newcomer to the realm of spatiotemporal data mining, I find the comprehensive resources at my disposal to be immensely valuable. However, I've encountered some challenges while attempting to reproduce the MTGNN, GraphWaveNet, and AGCRN models using the METR-LA/PEMSBAY dataset. Here is an outline of the problem I'm facing.

A ValueError emerged during the application of the inverse_transform method from the StandardScaler class within the scikit-learn library. The error message suggests that inverse_transform was provided with a three-dimensional array, while the method is designed to handle arrays of two dimensions or less.
https://github.com/user-attachments/assets/aee793b9-4cc5-499b-b0ba-2de6dc50be0e
The attached screenshot illustrates the error encountered while trying to reproduce the AGCRN model; similar errors were observed with MTGNN and GraphWaveNet as well.

Steps to Reproduce:
I have been following the steps outlined in the README to replicate the MTGNN, GraphWaveNet, and AGCRN models. Below are the commands I used for reproduction:

# To reproduce MTGNN, GraphWaveNet, and AGCRN on METR-LA
cd workMETRLA
python pred_AGCRN.py 0
python pred_GraphWaveNet.py 0
python pred_MTGNN.py 0

Additional Information:
Python Version: Python 3.8.10
torch Version: 2.0.0+cu118