Currently, I have my model trained with dimensions n x 10 x 6.
I want to test how my model performs when there is missing data in the features. I would like to know if it is possible to input some data as 1x10x3 to the previously trained model or if I could use 1x10x6 but with empty values.
Changing the input tensor size might raise errors, especially if you reduce the feature dimension and a particular layer expects its input to have a defined feature size (e.g.
You could set the “missing” features to some predefined values, but it depends on your use case what these values might be (there are no “empty” numbers and I would avoid using e.g.
NaN as a value).