Can I automate the training of the machine learning model with different parameters in AltAir PhysicsAI sotware (with same dataset) ?

Ashrut Sharma
Ashrut Sharma Altair Community Member
edited August 14 in Community Q&A

Hello,

I wanted to know if there is a way through which I can automate to change the parameters for training the model like  the epochs, depth, width, patience, learning rate with early stopping enabled for a particular stress contour. The automation should start training another model after one model is trained. I also want the lowest loss function value to be saved (or maybe the log file to be stored somewhere).

I want to find the best parameters so that I can use that in a much bigger dataset. 

 

Regards,

Ashrut Sharma

Tagged:

Answers

  • Garima_Singh
    Garima_Singh
    Altair Employee
    edited August 14

    Hi Ashrut,

    Thank you for posting the question on the Altair Community.

    We have demonstrated in a meeting to you the process of using HyperStudy for performing PhysicsAI's model hyper-parameter tuning.

    For the benefit of other customers who may be interested in this requirement. The solution details are as follows:

    Engineering Data Science (EDS) team has developed a custom HyperStudy Connector to perform DOE studies to understand the effect of the various PhysicsAI's model hyper-parameters on the model training time, model performance metrics such as loss value, validation loss value etc.

    Using HyperStudy's Pareto Plot result post-processing option, the effect of epochs on the PhysicsAI model training time can be studied (as shown below). This is one such example of using the custom HyperStudy Connector for the requirement.

    image

    If one has a similar requirement or have questions on the solution mentioned above, kindly send us a mail at hwsupport@altair.com & we will provide the required support.

    Many thanks.

    Kind Regards

    Garima Singh

Categories