Open vbast opened 1 year ago
This was a deliberate choice to make the interpretation easier for the development of the tool: i.e. all pixels are ML-predicted (incl the ones used for training). This choice can be revised if the tool is ready / made a user-defined option.
I've found that the restart files data provided as an input to task 4 (for training) is not identical in selected pixels to the restart files data generated as an output by task 4. In other words, the predicted data is different from training data in the pixels provided for that training. So, I am wondering what SPINacc does over the training pixels. Does it change it when writing back in the output restart? Shouldn't it write back exactly the same training data?