0.8.4

=====

- [NEW]: Improve performance when using pandas by reducing float array copies.

0.8.3

=====

- [NEW]: Update to TA-Lib C 0.8.3.

- [NEW]: This includes new TA functions: ALMA, ASI, BBW, CG, CHOP, CHOPTR,

CKSP, CRSI, CTI, DOSC, EMV, FISHER, FRAMA, IBS, ICHIMOKU, KST, KSTEXT,

KURTOSIS, KVO, MCGD, MEDIAN, PERCENTB, PSO, ROGERSSATCHELL, RVIR, SI, SQZMOM,

STC, SWAK_2PHP, SWAK_BP, SWAK_BUTTER, SWAK_GAUSS, SWAK_HP, VIDYA, WAVETREND,

ZIGZAG

- [NEW]: Additional moving average types ``MA_Type.VIDYA`` and

``MA_Type.ALMA``.

- [NEW]: Source builds download and verify TA-Lib C 0.8.3, build it as a

static library and link it into the extension. A compiler, CMake and network

access are required; a separate TA-Lib C installation is no longer needed.

Generated C files remain included in source distributions.

- [CHANGE]: Setting ``TA_INCLUDE_PATH`` or ``TA_LIBRARY_PATH`` uses an existing

TA-Lib C installation instead, without downloading or building the C library.

If only one is set, the other uses the standard search paths.

- [CHANGE]: Wheels statically link TA-Lib C through the same build process.

0.8.2

=====

- [NEW]: Support free-threaded CPython without enabling the GIL on import.

0.8.1

=====

- [CHANGE]: Native indicator calculations and stream operations release the

GIL on regular CPython too. Independent calls can run concurrently;

operations on a shared stream handle are protected by a per-handle mutex.

Python argument conversion, array allocation and result construction still

require Python thread state. Input arrays must not be modified or resized

during a call. Short calls can incur additional GIL handoff overhead.

- [CHANGE]: Changing process-wide unstable periods, compatibility or candle

settings while a call is active or any stream is alive now raises

``RuntimeError``. Settings cannot change between lookback calculation and

computation. Explicit shutdown also rejects active calls or streams.

- [FIX]: ``tools/threads_talib.py`` now benchmarks a fixed number of calls,

propagates worker failures and reports median throughput across thread

counts, replacing its racy shared counter.

- [CHANGE]: ``talib.stream`` is now the real streaming API of TA-Lib C 0.8.1:

``stream.SMA(close)`` returns a handle, not a value. ``handle.value`` is the

value at the last history bar, ``handle.update(bar)`` costs O(1) and returns

that bar's value, ``handle.peek(bar)`` evaluates a forming bar without

committing it, and ``handle.copy()`` forks it. ``stream.SMA.open_and_fill()``

returns the handle and the Function API's series in one pass. A multi-output

function answers with the same tuple the Function API returns. The old

last-value functions -- ``talib.stream.SMA``, ``talib.stream_SMA``, and their

``_ta_lib.pyi`` stubs -- are gone; ``talib/stream.pyi`` types the handles

instead.

Migrating is ``stream.X(...)`` -> ``stream.X(...).value``, and the compiler

cannot find the sites for you: ``if stream.CDLDOJI(o, h, l, c):`` used to test

the pattern and now tests a handle, which is always true.

- [NEW]: ``talib.InsufficientHistory``, raised when a stream is opened with

too little history. It is the library's one recoverable error, so it is

catchable on its own rather than as a bare ``Exception``.

- [FIX]: ``help(talib.SUPERTREND)`` and the ``abstract`` stub named the outputs

``real`` and ``integer``; they are ``supertrend`` and ``trend``, the names

``abstract.Function('SUPERTREND').output_names`` already reported.

- [FIX]: An empty array given to a function whose lookback is zero, such as

``talib.ACOS`` or ``talib.MA(x, timeperiod=1)``, made TA-Lib read and write

one element outside the buffers, which could crash the interpreter later or

corrupt memory silently. An empty input now returns empty outputs without

calling TA-Lib, whatever the function and its parameters. The bug dates from

0.4.27.

0.8.0

=====

- [NEW]: Support TA-Lib C 0.8.1, which is now the minimum required version.

- [NEW]: The 40 functions TA-Lib C added since 0.7.1: AC, ADR, AO, CMF, CMOU,

COPPOCK, CUMSUM, CVI, DONCHIAN, DPO, EFI, ER, ERI, FOSC, FRACTAL, HA, HMA,

KC, KDJ, MARKETFI, MASSI, NVI, PERCENTILE, PERCENTRANK, PVI, PVO, PVT,

QSTICK, RMA, RVI, RVOL, SMI, SUPERTREND, TSI, VHF, VORTEX, VWAP, VWMA, WAD,

ZLEMA.

- [NEW]: New moving averages: ``MA_Type.HMA``, ``MA_Type.DISABLED``,

``MA_Type.DEFAULT``, ``MA_Type.ZLEMA`` and ``MA_Type.RMA``.

- [NEW]: ``set_unstable_period()`` and ``get_unstable_period()`` accept

``'RMA'``, ``'HA'`` and ``'RVI'``.

- [FIX]: ``abstract`` raised ``KeyError`` on function and output flags added

after this wrapper was written; unknown flag bits are now ignored.

- [FIX]: A moving-average parameter not spelled exactly ``matype`` -- ``KDJ``'s

``slowk_matype`` -- defaulted to SMA rather than to the function's own

documented default.

- [FIX]: An ``integer`` output is documented as the candlestick -100/0/100

convention only for candlestick functions; ``SUPERTREND``'s is a trend

direction.

- [CHANGE]: ``APO`` and ``PPO`` now default ``matype`` to EMA, and ``BBANDS``

defaults ``timeperiod`` to 20, following TA-Lib C 0.8.1.

- [FIX]: ``set_unstable_period()`` targeted the wrong function for every id

after ``IMI`` (e.g. ``'RSI'`` set ``PLUS_DM``); ids now come from the C

header. ``'ADXR'``, ``'MFI'`` and ``'STOCHRSI'`` are no-ops that emit a

``DeprecationWarning``.

- [CHANGE]: Remove the unnecessary ``build`` runtime dependency and ``wheel``

build dependency.

- [NEW]: Upgrade to Cython 3.3.0

0.7.1

=====

- [FIX]: Fix wheels to build with TA-Lib C 0.7.1 properly.

0.7.0

=====

- [NEW]: Upgrade to Cython 3.2.8

- [FIX]: Adding ACCBANDS, AVGDEV, IMI to get_func_groups()

- [FIX]: Adding ACCBANDS, AVGDEV, IMI to docs pages

- [NEW]: Adding an ``abstract.pyi`` for typing

0.6.8

=====

- [FIX]: Remove accidental cython dependency.

- [NEW]: Upgrade to Cython 3.1.5

- [NEW]: Release binaries for Python 3.14

0.6.7

=====

- [NEW]: Upgrade to Cython 3.1.3

- [FIX]: Include .pxd files in the source .tar.gz

- [FIX]: Don't require pip, it's not needed

- [FIX]: Add back some import compat for backtrader

0.6.6

=====

- [FIX]: Silence project.license warnings when building.

- [FIX]: Fix regression in abstract function parameters not coercing integers

0.6.5

=====

- [FIX]: Support PEP-517 style installation

- [NEW]: Upgrade to Cython 3.1.2

- [NEW]: Build binary wheels for Windows, macOS, and Linux.

0.6.4

=====

- [FIX]: Adding typed for ACCBANDS, AVGDEV, IMI, and stream_* functions

- [FIX]: Fix for numpy 2.3.0, which removed the npy_1_7_deprecated_api

0.6.3

=====

- [FIX]: Fix issue building against numpy2

0.6.2

=====

- [NEW]: Adding indicators ACCBANDS, AVGDEV, IMI

0.6.1

=====

- [FIX]: Build properly against TA-Lib 0.6.2 and newer on Windows.

0.6.0

=====

- [FIX]: Make sure this wrapper works with TA-Lib 0.6.1 and newer releases.

0.5.5

=====

- [FIX]: Fix for numpy 2.3.0, which removed the npy_1_7_deprecated_api

0.5.4

=====

- [FIX]: Fix lib name on windows.

0.5.3

=====

- [BUG]: Make this release only support TA-Lib 0.4.0.

0.5.2

=====

- [NEW]: Provide py typed information for C wrapper

- [FIX]: Make sure this wrapper works with TA-Lib 0.6.1

0.5.1

=====

- [FIX]: include cython *.pxi files in the source tar.gz

0.5.0

=====

- [NEW]: Upgrade to Numpy 2.0

0.4.38

======

- [FIX]: Fix lib name on windows.

0.4.37

======

- [FIX]: Make sure to pin numpy<2 in the setup.py.

0.4.36

======

- [BUG]: Make this release only support TA-Lib 0.4.0

0.4.35

======

- [FIX]: Make sure this wrapper works with TA-Lib 0.6.1

0.4.34

=====

- [FIX]: include cython *.pxi files in the source tar.gz

0.4.33

======

- [NEW]: Upgrade to Cython 3.0.11

- [FIX]: Fix import of ``pandas`` and ``polars`` to throw underlying import errors

0.4.32

======

- [FIX]: Make sure numpy<2 is in the requirements.txt

0.4.31

======

- [FIX]: Fix build issue with numpy2 released by requiring numpy<2.0.0 for now

0.4.30

======

- [FIX]: Fix the C library imports to work again on Windows, oops.

0.4.29

======

- [NEW]: Upgrade to Cython 3.0.10

- [FIX]: Make TA_INCLUDE_PATH and TA_LIBRARY_PATH override default search list

- [FIX]: Silent some compiler warnings about const pointers

0.4.28

======

- [NEW]: Upgrade to Cython 3.0.0

- [FIX]: Use embedsignature to enable code completions?

0.4.27

======

- [NEW]: Upgrade to Cython 0.29.36

- [FIX]: Fix abstract info to keep values doubles when they are (e.g., nbdevup, nvdevdn)

- [FIX]: Don't throw exceptions when inputs are all NaN

0.4.26

======

- [NEW]: Upgrade to Cython 0.29.34

- [NEW]: Moved to https://github.com/ta-lib/ta-lib-python

- [FIX]: Fix deprecation warning for polars.DataFrame(columns=[...])

0.4.25

======

- [NEW]: Upgrade to Cython 0.29.32

- [FIX]: Allow Abstract API to be safely used with multithreading

0.4.24

======

- [FIX]: runtime_lib_dirs preventing install on Windows

0.4.23

======

- [NEW]: Upgrade to Cython 0.29.25

- [NEW]: Allow TA_INCLUDE_PATH and TA_LIBRARY_PATH to work on Windows.

- [FIX]: Remove pyproject.toml which was preventing some installs.

0.4.22

======

- [NEW]: Upgrade to Cython 0.29.24

- [NEW]: Support polars.Series in Function API

- [NEW]: Support polars.Series in Streaming API

- [NEW]: Support polars.DataFrame in Abstract API

- [NEW]: Switch tests to pytest from nose (no longer supported).

0.4.21

======

- [FIX]: Fixed {MIN/MAX/MINMAX}INDEX functions to return correct indices.

0.4.20

======

- [NEW]: Upgrade to Cython 0.29.23

- [FIX]: Fix Homebrew directory on Apple M1 devices

0.4.19

======

- [NEW]: Upgrade to Cython 0.29.20

- [NEW]: from talib import * (imports all func and stream)

- [FIX]: missing method signatures in PyCharm

0.4.18

======

- [NEW]: Upgrade to Cython 0.29.17

- [NEW]: Support TA_SetCompatibility

- [NEW]: Reduce size of generated C file and memory requirements to compile it.

- [FIX]: Fixed caching input price series in Abstract API

- [NEW]: Support pandas.Series in Streaming API

0.4.17

======

- [NEW]: Support pandas.Series in Function API

0.4.16

======

- [FIX]: Fix incorrect C header include on Windows.

0.4.15

======

- [FIX]: Import TA_FUNC_FLAGS, TA_INPUT_FLAGS, TA_OUTPUT_FLAGS into

talib.abstract to support backwards compatibility for backtrader

0.4.14

======

- [NEW]: Upgrade to Cython 0.27.3

- [NEW]: Add support for set/restore candle settings

- [DOC]: Note Function API only supports numpy.ndarray (not pandas.Series)

- [NEW]: Continuous integration using Travis on Python 2.7, 3.4, 3.5, 3.6

- [NEW]: Merge common, func, abstract, stream as pxi to _ta_lib.pyx

- [DOC]: Update documentation to warn about unstable functions

0.4.10

======

- [NEW]: Upgrade to Cython 0.23.4.

- [NEW]: Adding an experimental Streaming API (talib.stream).

- [NEW]: Support for "sunos"

- [FIX]: Fix issue initializing library for use with candlestick functions.

- [NEW]: Support for TA_SetUnstablePeriod and TA_GetUnstablePeriod

0.4.9

=====

- [NEW]: Support for python 2.6 and earlier for ordereddict.

- [FIX]: Add lib64 paths to setup.py.

- [NEW]: Upgrade to Cython 0.22.

- [FIX]: Allow running setup.py without numpy or cython

- [NEW]: Allow TA_LIBRARY_PATH and TA_INCLUDE_PATH environment variables.

0.4.8

=====

- [NEW]: Support pandas.Series and pandas.Dataframe inputs.

- [FIX]: Simple check to warn if ta-lib library is not installed

- [FIX]: Fix missing key error when inputs are not default names.

- [NEW]: Upgrade to Cython 0.20.

- [FIX]: Check all input array lengths are same.

- [FIX]: Check all input arrays for NaN elements.

0.4.7

=====

- [NEW]: Upgrade to Cython 0.19.1.

- [FIX]: Fix "periods" input arrays in abstract interface.

- [FIX]: Only require necessary input arrays in abstract interface.

0.4.6

=====

- [NEW]: Support for function_flags and output_flags properties in Functions

0.4.5

=====

- [NEW]: Better test coverage of abstract interface.

- [NEW]: Support for Python 3.

0.4.4

=====

- [FIX]: Make installation on Windows easier.

0.4.3

=====

- [NEW]: Upgrade to Cython 0.18.

- [FIX]: Improve docstrings for indicators in talib.func.

- [FIX]: Initialize and shutdown underlying TA-Lib only once each process.

This should give slight speedups when using many indicators (<5%).

- [DEPRECATED]: The old moving average types that were in talib.func are now

deprecated. See the next bullet item for the replacement, or you can

continue using them with "from talib import deprecated" for the time being.

- [NEW]: The new moving average types are in talib.MA_Type. It's a class with SMA,

EMA, DEMA, etc attribute variables you should now use. Human-readable lookups

are also possible by using MA_Type as a dict e.g.:

MA_Type[MA_Type.SMA] # returns "Simple Moving Average"

- [NEW]: In addition to the existing func interface, we now provide an

enhanced abstract interface too. Functions in "import talib" use the

original interface, or you can use "from talib import abstract" for

the full abstract interface. See the README, tools/example.py and

tests/abstract_test.py for more information. The full (commented)

source code is mostly located in talib/abstract.pyx. The basic usage is very

similar:

input_arrays = { 'open': np.random.random(100),

'high': np.random.random(100),

'low': np.random.random(100),

'close': np.random.random(100),

'volume': np.random.random(100) }

ema_20 = abstract.Function('ema', input_arrays, 20).outputs # function names not case-sensitive

slowd, slowk = abstract.Function('STOCH')(input_arrays, 15, 3, 0, 3, 0)

upper, middle, lower = abstract.Function('bbands')(input_arrays, timeperiod=20)

0.4.2

=====

- Move the functions into ``talib.func``, to prepare for ``talib.abstract``

in a future release.

0.4.1

=====

- Support for MacPorts.

- Fix for FreeBSD.

0.4.0

=====

First version.