Pandas comparison

Coral uses pandas as a reference for supported operations. These conventions matter when comparing results:

Area	Coral behavior
Grouping and value_counts	Keys keep first-seen order. Pandas normally sorts group keys or sorts value counts by frequency.
outer_join	Left rows remain in order, followed by right-only rows. Its key column is converted to strings.
Joins with overlapping names	A right non-key column receives the _right suffix.
Join keys	As in pandas merge, duplicate keys give every matching pair and missing keys (NaN, or masked integers) match each other.
melt	Output columns are variable, value, then id columns; rows are grouped by source value column.
Missing values	Float columns use NaN; integer columns have a separate bool mask; string and bool columns have no missing marker.
GroupBy on NaN or masked keys	Missing keys form one group. Pandas drops them by default (dropna=True).
GroupBy aggregations	Float NaNs are not skipped; masked integers are. Pandas skips both.
describe	Numeric columns use sample standard deviation (ddof=1) and linear-interpolation quartiles, as pandas does; non-numeric columns are omitted.
sort_by on strings	Coral's own character ranking, lowercase before uppercase. Pandas compares code points, uppercase first.
ewm	adjust=False form only.

See I/O for file formats and output limits, including integer masks that are not preserved by the writers. See limitations for column-type constraints and the functions that chelis build rejects.