Describe

Coral.Frame.describe(frame) summarizes numeric columns and skips string and bool columns. The result has a stat string column with eight rows: count, mean, std, min, 25%, 50%, 75%, and max. Each float or integer input column becomes a FloatCol of results under the same name, in the input's column order.

NaNs in float columns and masked values in integer columns are left out before any statistic is computed, so count is the number of valid values. Standard deviation uses the sample divisor (ddof=1). Integer values are converted to f32.

Infinities are not left out. They follow IEEE arithmetic: mean becomes inf or NaN, std becomes NaN, and a quartile that interpolates against an infinity becomes inf or NaN. For [1.0, inf, 3.0] the result is [3.0, inf, NaN, 1.0, 2.0, NaN, inf, inf]. Replace infinities before calling describe if you need finite statistics.

The quartiles use linear interpolation, the pandas default: for quantile q over the c sorted valid values, the position is q * (c - 1), and a fractional position interpolates between its two neighbors. For [1.0, 2.0, 3.0, 4.0], 25%, 50%, and 75% are 1.75, 2.5, and 3.25, and std is 1.2909944.

module Coral.BookDescribe
import Coral.Frame (FloatCol, from_pairs, describe, get_float_col)
export (main)
def main() -> f32 = {
  frame = from_pairs([("price", FloatCol(to_tensor([10.0f32, div(0.0f32, 0.0f32), 40.0f32])))])
  summary = describe(frame)
  index(to_list(get_float_col(summary, "price")), 0i64)
}

The first result is the count of non-NaN prices, so main returns 2.0.

Small and empty columns

Valid values	Result column
None (all NaN or all masked)	count is 0.0; the other seven rows are NaN
One	std is NaN; mean, min, the quartiles, and max equal that value
Two or more	Every row is a number

For a column holding 5.0 and three NaNs, the result is [1.0, 5.0, NaN, 5.0, 5.0, 5.0, 5.0, 5.0].

describe is one of the functions that chelis build rejects; see limitations.