Construction

A Frame stores columns of one row count. from_pairs keeps the listed column order and rejects duplicate names or unequal column lengths. The four column forms are:

Form	Stored values
FloatCol	tensor[n, f32]
IntCol	tensor[n, i64] and a tensor[n, bool] missing-value mask
BoolCol	tensor[n, bool]
StringCol	List[string]

Use int_col_of_list to make an integer column with no missing entries.

Function	Signature	Failure
from_pairs	(pairs: List[(string, Column[n])]) -> Frame[n]	from_pairs: duplicate column name, from_pairs: mismatched column lengths
from_columns	(cols: Dict[string, Column[n]]) -> Frame[n]	Same as from_pairs
empty	(schema: Dict[string, ColumnType]) -> Frame[n]	None
int_col_of_list	(values: List[i64]) -> Column[n]	None

from_pairs([]) returns a frame with no columns and zero rows.

from_columns and empty take their column order from the dictionary's entry order, as dict_entries returns it; use from_pairs when you need to state the order explicitly. empty makes a frame with the named columns and zero rows; the ColumnType values are IntType, FloatType, StringType, and BoolType.

module Coral.BookConstruct
import Coral.Frame (FloatCol, StringCol, from_pairs, get_float_col)
export (main)
def main() -> f32 = {
  frame = from_pairs([("price", FloatCol(to_tensor([1.0f32, 2.0f32]))), ("name", StringCol(["a", "b"]))])
  prices = get_float_col(frame, "price")
  index(to_list(prices), 1i64)
}

get_float_col returns a Chelis tensor, so main reads the second price and returns 2.0.

module Coral.BookEmpty
import Coral.Frame (IntType, StringType, empty, columns, nrows)
export (main)
def main() -> (List[string], i64) = {
  frame = empty(dict_of([("id", IntType), ("city", StringType)]))
  (columns(frame), nrows(frame))
}

main returns [id, city] and 0.

Reading a frame

Function	Signature	Failure
get_float_col	(df: Frame[n], name: string) -> tensor[n, f32]	missing column: NAME, column is not float: NAME
get_int_col	(df: Frame[n], name: string) -> tensor[n, i64]	missing column: NAME, column is not int: NAME
get_bool_col	(df: Frame[n], name: string) -> tensor[n, bool]	missing column: NAME, column is not bool: NAME
get_string_col	(df: Frame[n], name: string) -> List[string]	missing column: NAME, column is not string: NAME
get_column	(df: Frame[n], name: string) -> Column[n]	missing column: NAME
column_type	(df: Frame[n], name: string) -> ColumnType	missing column: NAME
columns	(df: Frame[n]) -> List[string]	None
nrows, ncols	(df: Frame[n]) -> i64	None

A missing name or a mismatched column type fails when the accessor runs. get_int_col returns only the values; a missing entry reads as its stored number, usually 0. Read the mask with is_nan_col (see Missing values) or match on get_column to get both tensors of an IntCol.