The whole point of the exercise was to have the computer find rules on its own: If A and B, then C. E.g. let’s say we have a dress that is also orange and made of light cotton, most likely it is going to be a summer dress. While these logical constructs can seem quite obvious and simple to the trained eye, the computer has no way to recognize them without previous knowledge, enter Kohonen Self-organizing maps.
Kohonen maps convert highly dimensional spaces into simple bidimensional maps, we can reduce as many attributes or dimensions as we want into simple X and Y axes.
Making use of the connections between various fashion articles we were able to generate a Kohonen map, allowing us to easily visualize and extract connections between categories for future automated categorization.