ScatterPlot Class Too few legends / colors exception

[haifengl]

ScatterPlot Class Too few legends / colors exception

[haifengl]

Sorry, this is a documentation problem. The labels should be in the range [0, k), where k is the number of classes. In your sample code, we should have

val exampleLabels = Array(0, 1)

We could change Array to Map. However, there is a performance issue for large number of points. Some people tried to plot millions data points. Map will make the UI unusable.

[Xyclade]

Oh maybe I explained it wrongly:

My suggestion was to keep the array for the labels the same, but to change the implementation in the constructor, such that next to the array you get a global lookup table which holds the unique values and their Colors / legends. I'll work it out, such that you can see exactly what I mean, because explaining it is rather difficult.

Ps. It should not affect the performance, as the extra computation is only n where n is the amount of labels. At least, in the way I see the adjustment.

[haifengl]

The performance is not in the constructor but in plot. For each data point, we may need change color or legend, now it is done by a Map lookup instead of array index. In terms of complexity, it is still O(n) for HashMap. But the constant factor of n is much bigger now. In your experiments, please try millions data points to make sure if it is still ok. Maybe I am wrong. Thanks!

[Xyclade]

I will test with a lot of datapoints. This is the fix I'm suggesting. Since the lookup in the hashmap should be O(1) it should not affect the plot method in performance.

In a second commit I merged the two for loops as I missed the second one in the first commit.
I'll let you know how the test goes.

[Xyclade]

For 10 million datapoints the time to plot was equal (+/- 1 second on 1.04 minutes plotting time) for both methods.

[haifengl]

Cool. BTW, we don't need to check if id is negative now. Can you rename valueLookupTable to classLookupTable? Thanks!

I had a commit updating the comment. Can you please update the comment again? The API may be confusing now. The user may not have clear idea of what is the color/legend of class 100.

[Xyclade]

I removed the negative check and renamed the table. However I cannot find the commit regarding the comment, could you point me out where I should update this comment? 👍

[haifengl]

Sorry, forgot to push to github. It is there now. Thanks!

[Xyclade]

Tnx for merging it 😄