Learning documentation
I consult the documentation at http://osr-cesbio.ups-tlse.fr/echangeswww/TheiaOSO/iota2_documentation/master/i2_classification_tutorial.html#output-tree-structure
A note : The learningSamples directory does not contain shapefiles but sql files corresponding to pixel numbers.
My questions are about the contents of the formatingVectors and SampleSelection directories which contain shapefiles.
Iota2 is run in a configuration using multiple runs (configuration file attached). fakaraTest.cfg
The input file defined by ground_truth contains 444 polygons.
The output files in the sampleSelection directory contain shapefiles describing 308 polygons. These are not the same polygons depending on the file (samples_region_1_seed_0.dbf, ...seed_1..., ...seed_2...). Each of these files is composed of 7 columns, 3 of which are named seed_0, seed_1, seed_2, taking as value "learn" or "validation". The number of these columns (3) corresponds to the number of runs.
The output directory formattingVectors contains two shapefiles corresponding to the two processed tiles. Their respective number of polygons is 305 and 86 (305 +86 = 391). Each of these files is composed of 7 columns, 3 of which are named seed_0, seed_1, seed_2, taking as value "learn" or "validation". The number of these columns (3) corresponds to the number of runs.
- This directory contains the learning samples for each tile. What does learning mean: training, validation, training + validation?
- Why aren't all the polygons in the ground_truth file dispatched in the formatingVectors directory?
- Does the SampleSelection directory contain the training samples?
- What are the columns named seed_0, seed_1, seed_2 in the sampleSelection directory files?
- What are the columns named seed_0, seed_1 , seed_ 2 in the formattingVectors directory files?
- Is the distribution of samples between training and validation areas done at each run? The documentation says "run : Number of independant runs processed. Each run has its own learning samples... ". If the traning/validation split of the samples is done for each run, how to make it identical for all the runs ?
Thank you for your help,
Have a nice day,
Yves