Skip to content
Chelis docs

Reshape

Coral.Reshape provides wide-to-long and long-to-wide transformations on Frame values. All four operations are compile-checked against chelis v0.7.6.

Validated operations in the current slice:

  • melt: wide to long, repeat id columns, pivot value columns into variable/value rows
  • pivot: long to wide, unique values of columns_col become new column names
  • stack: thin shim, melt with no id columns, all columns as value columns
  • unstack: thin shim, pivot on a stacked frame given an explicit index column

v0.1 constraints:

  • values_col for pivot must be FloatCol; other types fail with a runtime error.
  • value_cols for melt must all be FloatCol.
  • index_col and columns_col for pivot must be StringCol.
  • id_cols for melt must be StringCol.
module Coral.BookReshape
import Coral.Frame (from_pairs, nrows, ncols)
import Coral.Reshape (melt)
export (main)
def main() -> int64 = {
  frame = from_pairs([("city", StringCol(["london", "paris", "oslo"])), ("qty", FloatCol(to_tensor([cast(5.0, f32), cast(6.0, f32), cast(7.0, f32)]))), ("price", FloatCol(to_tensor([cast(10.0, f32), cast(20.0, f32), cast(30.0, f32)])))])
  melted = melt(frame, ["city"], ["qty", "price"])
  nrows(melted)
}

The output frame has columns variable, value, and city with nrows = 3 * 2 = 6 (one row per original row per value column).

The gate today is fixture-backed plus compile-checked. A full executed runtime parity lane for Reshape will follow when the bare-build harness is extended to cover multi-column frames.