1. About StaticFrame¶
Prior to the 1.0 release, interfaces may change. Please assist in development by reporting any bugs or request any missing features.
StaticFrame is not a drop-in replacement for Pandas. While some conventions and API components are directly borrowed from Pandas, some are completely different, either by necessity (due to the immutable data model) or by choice (offering more uniform, less redundant, and more explicit interfaces). As StaticFrame does not support in-place mutation, architectures that made significant use of mutability in Pandas will require refactoring.
StaticFrame is lightweight. It has few dependencies (Pandas is not a dependency).
StaticFrame does not aspire to be an all-in-one framework for all aspects of data processing and visualization. StaticFrame focuses on providing efficient and powerful data structures with consistent, clear, and stable interfaces.
StaticFrame aspires to have comparable or better performance than Pandas. While this is already the case for some core operations (See Performance), some important functions are far more performant in Pandas (such as reading delimited text files via
pd.read_csv). StaticFrame provides easy conversion to and from Pandas to bridge needed functionality or performance.
StaticFrame relyies entirely on NumPy for types and numeric computation routines. NumPy offers desirable stability in performance and interface. For working with SciPy and related tools, StaticFrame exposes easy access to NumPy arrays, conversion to and from Pandas and Arrow, and support for reading from and writing to a wide variety of storage formats.
static_frame.Frame store data in immutable NumPy arrays. Once created, array values cannot be changed. StaticFrame manages NumPy arrays, setting the
ndarray.flags.writeable attribute to False on all managed and returned NumPy arrays.
>>> import static_frame as sf >>> import numpy as np >>> s = sf.Series((67, 62, 27, 14), index=('Jupiter', 'Saturn', 'Uranus', 'Neptune'), dtype=np.int64) >>> s #doctest: +NORMALIZE_WHITESPACE <Series> <Index> Jupiter 67 Saturn 62 Uranus 27 Neptune 14 <<U7> <int64> >>> s['Jupiter'] = 68 Traceback (most recent call last): File "<stdin>", line 1, in <module> TypeError: 'Series' object does not support item assignment >>> s.iloc = 68 Traceback (most recent call last): File "<stdin>", line 1, in <module> TypeError: 'InterfaceGetItem' object does not support item assignment >>> s.values = 68 Traceback (most recent call last): File "<stdin>", line 1, in <module> ValueError: assignment destination is read-only
To mutate values in a
Frame, a copy must be made. Convenient functional interfaces to assign to a copy are provided, using conventions familiar to NumPy and Pandas users.
>>> s.assign['Jupiter'](69) <Series> <Index> Jupiter 69 Saturn 62 Uranus 27 Neptune 14 <<U7> <int64> >>> s.assign['Uranus':](s['Uranus':] - 2) <Series> <Index> Jupiter 67 Saturn 62 Uranus 25 Neptune 12 <<U7> <int64> >>> s.assign.iloc[[0, 3]]((68, 11)) <Series> <Index> Jupiter 68 Saturn 62 Uranus 27 Neptune 11 <<U7> <int64>
Immutable data has the overwhelming benefit of providing the confidence that a client of a
Frame cannot mutate its data. This removes the need for many unnecessary copies, and forces clients to only make copies when absolutely necessary.
There is no guarantee that using immutable data will produce correct code or more resilient and robust libraries. It is true, however, that using immutable data removes countless opportunities for introducing flaws in data processing routines and libraries.
The ideas behind StaticFrame developed out of years of work with Pandas and related tabular data structures by the Investment Systems team at Research Affiliates, LLC. In May of 2017 Christopher Ariza proposed the basic model to the Investment Systems team and began implementation. The first public release was in May 2018.
The following presentations and interviews describe StaticFrame in greater depth.
PyData LA 2019: “The Best Defense is not a Defensive Copy” (lightning talk starting at 18:25): https://youtu.be/_WXMs8o9Gdw
PyData LA 2019: “Fitting Many Dimensions into One The Promise of Hierarchical Indices for Data Beyond Two Dimensions”: https://youtu.be/xX8tXSNDpmE
PyCon US 2019: “A Less Kind, Less Gentle DataFrame” (lightning talk starting at 53:00): https://pyvideo.org/pycon-us-2019/friday-lightning-talksbreak-pycon-2019.html
Talk Python to Me, interview: https://talkpython.fm/episodes/show/204/staticframe-like-pandas-but-safer
PyData LA 2018: “StaticFrame: An Immutable Alternative to Pandas”: https://pyvideo.org/pydata-la-2018/staticframe-an-immutable-alternative-to-pandas.html
These members of the Investment Systems team have contributed greatly to the design of StaticFrame:
Thanks also for additional contributions from GitHub users: