search

VH-Lab / NDIcalc-vis-matlab

NDI calculator objects and documents for vision research
Other
1 stars 1 forks source link

Errors found when comparing mock contrast tuning docs #25

Closed aviL221 closed 3 months ago

aviL221 commented 3 months ago

I ran the following on the command line: [b,errormsg,b_expected,doc_output,doc_expected_output] = contrast_tuning_calc.test('standard',1,1); While b_expected was 1, b was 0. Errormsg contained 17 errors. Most were of the variety "Arrays of insert_field are not the same size". I compared doc_output and doc_expected_output by comparing the fields of doc_output.document_properties.contrast_tuning and doc_expected_output.document_properties.contrast_tuning. The dimensions of fields in doc_output.document_properties.contrast_tuning.fit and doc_expected_output.document_properties.contrast_tuning.fit are flipped - that is, the row vectors in one are column vectors in the other, and vice versa. This seems to be interpreted by ndi.calc.vis.contrast_tuning/compare_mock_docs as having arrays of different size. The dimensions of fields in doc_output.document_properties.contrast_tuning.tuning_curve and doc_expected_output.document_properties.contrast_tuning.tuning_curve are also flipped - except for one field: doc_output.document_properties.contrast_tuning.tuning_curve.individual is a 1x10 cell array, while doc_expected_output.document_properties.contrast_tuning.tuning_curve.individual is a 10x10 array of type double. Hence, these are also interpreted as arrays of different sizes.

stevevanhooser commented 3 months ago

Yes, I've encountered this error as I've made other calculators.

The issue arises from the fact that when documents are converted to JSON and back, vectors that are not character arrays are always turned into columns. The best solution is to actually edit the calculator so that vectors are made as columns, so there is no difference between how they are at creation and how they come back after being read from the database or JSON file.

You can use vlt.data.column(v) to convert a vector v into a column (or leave it unchanged if it is already a column). One just has to make sure not to treat matrixes this way (or they will be smooshed into a column). So only the vector data should be put through vlt.data.column.

stevevanhooser commented 3 months ago

When you fix it, you can close the issue. You can also reference the issue in your commit by saying something like:

commit -am "changed vectors to columns in output of contrast_calc; closes #25"