- Configuration settings in the editable fields:
- Number of Epochs: Determines how many times to train over the entire dataset
- Batch Size: Dictates how many samples to process at once. Limited by GPU DRAM.
- Learning Rate: Determines the step size to train the model. Too low a number of trains slowly and too high a value would overshoot the optimal solution and perform poorly.
- Training configuration recommended values:
- **25 epochs is recommended if training with 3 or more classes (Not including Unknown)
- The T-SDK makes use of the machine’s multiple CPU cores to train at a faster rate. The user can adjust CPU usage through the cpu_cores_train parameter. The default value to this parameter is cpu_cores_train = 2, which utilizes 50% of the CPU. Using higher values of cpu_cores_train may induce T-SDK instability
- Training seeds: Set training_seed1, training_seed2, and training_seed3 from value 0 to 999, recommended setting to 0
- Initialize tab (optional): Select initialized weights to start a new training
