Bug about plotting the band in DeePTB-e3

[hwf2002]

Details

when i use the "override_overlap" to draw the band in deeptb-e3, I get a warning like that:
dargs.dargs.ArgumentKeyError: [at location task_options] undefined key override_overlap is not allowed in strict mode
and if I do not use the overlaps from DFT, it says the S(k) are not positive-definite, though i have trained 2000+ epoch and get the train_lost in 0.0028.
so what's the problems?

[hwf2002]

and my DeePTB version is 2.2.1.dev4+f9c4d66

[AsymmetryChou]

Thank you for your bug reportting. Can you paste the details of the bug outputting?

[Lonya0]

Thank you for your report. The issue is clearly articulated: your model has not been trained sufficiently, resulting in a non-positive definite Overlap matrix. However, it is worth noting that, for certain structural models, achieving adequate training to generate a positive definite Overlap matrix is extremely challenging. To address this, we have introduced a feature that allows the input of Overlap matrices generated by Density Functional Theory (DFT). We recommend using an external DFT tool to generate an Overlap matrix and employing the override_overlap function. Alternatively, you may continue training the model, though it should be noted that in some cases, prolonged training may still not yield a positive definite predicted Overlap matrix.

[hwf2002]

Thank you for your report. The issue is clearly articulated: your model has not been trained sufficiently, resulting in a non-positive definite Overlap matrix. However, it is worth noting that, for certain structural models, achieving adequate training to generate a positive definite Overlap matrix is extremely challenging. To address this, we have introduced a feature that allows the input of Overlap matrices generated by Density Functional Theory (DFT). We recommend using an external DFT tool to generate an Overlap matrix and employing the override_overlap function. Alternatively, you may continue training the model, though it should be noted that in some cases, prolonged training may still not yield a positive definite predicted Overlap matrix.

yes. As you said, I used the override_overlap to plot the band, and the problem is said above before.and the picture shows the full warning information.

[Lonya0]

Thank you for your report. The issue is clearly articulated: your model has not been trained sufficiently, resulting in a non-positive definite Overlap matrix. However, it is worth noting that, for certain structural models, achieving adequate training to generate a positive definite Overlap matrix is extremely challenging. To address this, we have introduced a feature that allows the input of Overlap matrices generated by Density Functional Theory (DFT). We recommend using an external DFT tool to generate an Overlap matrix and employing the override_overlap function. Alternatively, you may continue training the model, though it should be noted that in some cases, prolonged training may still not yield a positive definite predicted Overlap matrix.

yes. As you said, I used the override_overlap to plot the band, and the problem is said above before.and the picture shows the full warning information.

have you updated the package? the feature was recently updated.

[Lonya0]

*recently (was updated two monthes ago)