Currently the batch size can have a minimum value of 2 and a maximum value of 7. The minimum value of 2 is due to how the fits are postprocessed. When a batch size of 1 is used it changes the shape of the weights dictionary causing removing an index. This will have to be resolved in order to allow for batch size of 1. The maximum batch size of 7 is set by memory constraints in calculating the hessian. When a larger batch size is used the hessian call in postprocess.recalculate_with_chosen_weights returns an OOM error.
Currently the batch size can have a minimum value of 2 and a maximum value of 7. The minimum value of 2 is due to how the fits are postprocessed. When a batch size of 1 is used it changes the shape of the weights dictionary causing removing an index. This will have to be resolved in order to allow for batch size of 1. The maximum batch size of 7 is set by memory constraints in calculating the hessian. When a larger batch size is used the hessian call in postprocess.recalculate_with_chosen_weights returns an OOM error.