Author Topic: configuration for Training set created by random displacement parameters  (Read 2438 times)

0 Members and 1 Guest are viewing this topic.

Offline pshinyeong

  • Heavy QuantumATK user
  • ***
  • Posts: 38
  • Country: kr
  • Reputation: 0
    • View Profile
Hello,

I have successfully used the Defect Training Workflow template to obtain the Moment Tensor Potential (MTP). I would now like to review the structures used in the training set, specifically those generated by applying the Rattle Atom, Rattle Cell, and Rattle Volume random displacement parameters.

While I can see that these random displacement parameter objects have been added to the table, I am unsure how to visually inspect or view them. Could you please guide me on how to access or visualize the specific configurations generated by these displacement parameters?

Thank you in advance for your assistance.

Offline Anders Blom

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 5583
  • Country: dk
  • Reputation: 100
    • View Profile
    • QuantumATK at Synopsys
The tables are not for visualization, and in fact often do not contain any configurations, only the instructions how to make them. But you should have a file that contains a TrainingSet. If you select this you can see the training configurations in the Preview window on the right. Also, and perhaps better, there should be a file containing either an "mtp" or "mtp_active_learning" object. If you right-click these, they can be opened in the Movie Tool where you can see all configurations.

Offline pshinyeong

  • Heavy QuantumATK user
  • ***
  • Posts: 38
  • Country: kr
  • Reputation: 0
    • View Profile
Thank you for your response.

After running the Defect Training Workflow, I was unable to locate any file containing the Training Set. However, I do have the 'mtp' object, which I can open in the Movie Tool. In my case, two defect configurations were generated by the defect block, and I used a sample size of 10, 20, and 10 for Rattle Atom, Rattle Cell, and Rattle Volume parameters, respectively.

I am curious about the resulting trajectory, as it consists of 140 steps, whereas I was expecting 80 steps (40 configurations per defect). Could you help clarify why this might be the case?

Thank you for your assistance.


Offline Anders Blom

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 5583
  • Country: dk
  • Reputation: 100
    • View Profile
    • QuantumATK at Synopsys
The algorithm is complex, but these numbers are not going to be exactly obeyed, more of a guidance to the size you want. The are many different combinations of cell distortions and moving atoms going on in the internal machinery to get as good sampling as possible.