BME AI-Studio Training Stuck at 821/1024 Training Rounds

[AntoineSV]

I've been slowly collecting data for a regression model I'm training, I have about 90 specimens with varying target values totaling 60h. I've been getting better results with low batch size (4) and 1024 training rounds with previous iterations of my model, but this time around (now that I've added about 56 speciemens) it stayed stuck at 821/1024. I've tried it 3 times and on two computers and it always stalled at the 821 mark.

I've tried lowering the number of round to 512 which worked, and raising the batch size to 16 while keeping the number of rounds to 1024 also worked.

So there seems to be an issue with larger runs.

I'm on Linux.

[BSTRobin]

Hi AntoineSV,

Thanks for your inquiry.
BME AI-Studio provides configuration options for training parameters, and different gases and training parameters result in different training results. Usually, this requires selecting the optimal training parameters based on the actual gas type and testing environment, and the training results.

[AntoineSV]

I'm reporting a bug. The program hangs during training. It stays stuck at a training round and just keeps loading. This is a reccuring issue and on both Windows and Linux with sufficently big enough dataset. It's preventing me from properly training models.