Why do I get the feeling that you could programmatically name all these colors in some trivial way? Then you wouldn't end up with 30x "Blue" in your dataset.
Let's imagine we take the dominant byte of each color only. That's 4096 different color combinations. The lower byte makes up another 4096 color variations to each of those dominant colors. Just name 4096 "Color" words, like "Blue", "Aquamarine" etc. Then 4096 "useless adjectives" like "Mindful" and "Exotic".
And viola:
Exotic Blue - 0x0xFx | xAx3x5
Mindful Blue - 0x0xFx | x5x5x5
etc.
8000 words is something a single person could do by themselves, and it would generate all 16.7 million colors. You could even have some intuitive understanding. "Aha, Mindful Blue is a little bit grayer than just 'Blue'." This isn't even the most useful schema you could come up with...
Realistically, it's foolish to even do this in the RGB colorspace. Humans are not going to notice the difference between #000000 and #000001. Colors should be organized in a way that's useful. Hierarchical color naming at least gets you partway there.
Let's imagine we take the dominant byte of each color only. That's 4096 different color combinations. The lower byte makes up another 4096 color variations to each of those dominant colors. Just name 4096 "Color" words, like "Blue", "Aquamarine" etc. Then 4096 "useless adjectives" like "Mindful" and "Exotic".
And viola: Exotic Blue - 0x0xFx | xAx3x5 Mindful Blue - 0x0xFx | x5x5x5 etc.
8000 words is something a single person could do by themselves, and it would generate all 16.7 million colors. You could even have some intuitive understanding. "Aha, Mindful Blue is a little bit grayer than just 'Blue'." This isn't even the most useful schema you could come up with...
Realistically, it's foolish to even do this in the RGB colorspace. Humans are not going to notice the difference between #000000 and #000001. Colors should be organized in a way that's useful. Hierarchical color naming at least gets you partway there.