There are many ways to style a data layer by a numeric field. For example, a Continuous style maps a color (or point size) for every unique value in your data, similarly to a categorical visualization. In this case, the color for each value is interpolated from the endpoint colors.

However, it’s more common to group values from numeric fields into classes (green from 0 to 5, yellow from 6 to 10, etc). Classes can be chosen manually or picked from one of Felt’s default classification methods:

• Equal Intervals: classes are split by an interval of constant value (0, 5, 10, 15…). Useful for evenly distributed data. Equal interval classifications are very sensible to outliers - very small or large values can lead to the majority of values to be grouped into just one or two classes. This classification method follows an Arithmetic Series, where the value of each breakpoint is equal to the previous value plus a constant value.

• Quantiles: contrary to equal intervals, the quantiles classification ensures that every class contains the same number of values. This can look nice from a visual standpoint, but often quantiles classifications can be misleading, since the size of each class may vary widely. They work well for evenly distributed datasets.

• Standard Deviation: the Standard Deviation method chooses classes based on the number of standard deviations from the mean. This classification method works very well for data that follows a bell curve (a Gaussian or “normal” distribution).

By default, the legend in a Standard Deviation classification does not show the breakpoint values, using instead the nº of standard deviations.

• Geometric Intervals: a classification method where the size of each class increases exponentially, very useful for right-skewed data. The term “Geometric” comes from the fact that the class breaks are chosen from a geometric series (where each value equals the previous value, multiplied by a constant coefficient).

• Jenks Natural Breaks: the Jenks optimization method can be used to find “natural” classes inside your data by minimizing the variation inside classes while maximizing the variation between classes. It works well for most kinds of distributions, which is why it’s the default method in Felt.