Data Structures

• Only DataFrames

• No Series

Data Types

• Only primitive types - int, float, boolean, numpy.unicode

• No object data types

Row and Column Labels

• No index, meaning no row labels

• No hierarchical index

• Column names must be strings

• Column names must be unique

• Columns stored in a numpy array

Subset Selection

• Only one way to select data - [ ]

• Subset selection will be explicit and necessitate both rows and columns

• Rows will be selected only by integer location

• Columns will be selected by either label or integer location. Since columns must be strings, this will not be amibguous

• Column names cannot be duplicated

All selections and operations copy

• All selections and operations provide new copies of the data

• This will avoid any chained indexing confusion

Development

• Must use type hints

• Must use 3.6 - fstrings

• Must have numpy, bottleneck, numexpr

Small feature set

• Implement as few attributes and methods as possible

• Focus on good idiomatic cookbook examples for doing more complex tasks

Only Scalar Data Types

No complex Python data types - [x] bool - always 8 bits, not-null - [x] int - always 64 bits, not-null - [x] float - always 64 bits, nulls allowed - [x] str - A python unicode object, nulls allowed - [ ] categorical - [ ] datetime - [ ] timedelta