the predict.py doesn't contain inverse design

[zangowzidane]

Thank you for sharing this impressive code! Seems the code in predict.py does not work for inverse design from specification but works for predicting the S-parameters and gain with certain antenna parameters. So does it mean the predict.py is just for calling the trained function? if this understanding is correct, why do the 6 parameters need a range？

[youxch]

Thank you for your question.

Question 1

Yes, predict.py is used to predict the S-parameters and gain of the U-shaped patch antenna. A total of 2 million samples of structures with their S-parameters and gain are predicted, so inverse-design targets can actually be quickly searched within these 2 million samples (tested to be applicable to most situations).

Question 2

The parameter ranges of the training set and the prediction set are different. The prediction set has a denser and wider range of parameters compared to the training set. After testing, we have found that the predicted samples within the given range are accurate and reliable.

I hope I have answered your questions. If you have any further doubts or queries, please feel free to share them.