TaCuda — Python Bindings
Version 0.9.3 covers the 237 TA-Lib 0.8.3 functions with matching
native libraries and bindings. GPU RSI uses double-precision Wilder state.
The Python and Node.js loaders validate explicit native-library overrides.
See the compatibility guide.
New streaming families are a separate API surface. Third-party formula
licenses ship in THIRD_PARTY_NOTICES.txt.
Python package for the native tacuda CUDA library.
TA-Lib 0.8.3 coverage (237 functions): indicators, MACDEXT, 19 vector math kernels,
candlestick patterns, plus typed core APIs with runtime dtype selection and the
zero-copy device API.
Prerequisites
- NVIDIA GPU (Compute Capability >= 6.0)
- For official binaries: a driver compatible with CUDA 12.9; source builds use the toolkit selected by CMake
- The Windows x64 and Linux x86-64 wheels include the native library. Source installations need a matching native build.
- Python >= 3.8, NumPy >= 1.20
Package Scope
- High-level NumPy wrappers for indicator functions
- Precision-aware core indicators (
sma,ema,rsi,macd,atr,bbands) viadtype="float16|float32|float64" OHLCVconvenience container and input validation helpers
Installation
pip install tacuda
On Windows the wheel locates its native DLL and uses CUDA_PATH\bin for the
CUDA runtime. Set TACUDA_LIBRARY to a full path to select a different build.
Quick Start
import numpy as np
from tacuda import sma, ema, rsi, macd, bbands, OHLCV
# Generate sample data
prices = np.random.randn(100_000).cumsum().astype(np.float32) + 100
# Moving averages
sma20 = sma(prices, 20)
ema12 = ema(prices, 12)
# RSI
rsi14 = rsi(prices, 14)
# MACD (returns macd_line, signal, histogram)
macd_line, signal, hist = macd(prices, 12, 26, 9)
# Bollinger Bands (returns upper, middle, lower)
upper, middle, lower = bbands(prices, 20, 2.0, 2.0)
# OHLCV container
ohlcv = OHLCV.from_columns(open_arr, high_arr, low_arr, close_arr, volume_arr)
MACDEXT and vector math
import tacuda
# Independent MA types per line (0=SMA 1=EMA 2=WMA ... 8=T3)
ext_macd, ext_signal, ext_hist = tacuda.macdext(
prices, fastPeriod=12, fastType=1, slowPeriod=26, slowType=0,
signalPeriod=9, signalType=2)
# Element-wise GPU kernels
log_prices = tacuda.ln(prices)
spread = tacuda.sub(high_arr, low_arr)
Zero-copy device pipeline
n = len(prices)
d_in = tacuda.device_alloc(n)
d_out = tacuda.device_alloc(n)
try:
tacuda.device_upload(prices, d_in) # 1 PCIe transfer in
tacuda.sma_d(d_in, d_out, n, 20) # compute on GPU
tacuda.sin_d(d_out, d_out, n) # chain kernels, zero copies
tacuda.device_sync()
result = tacuda.device_download(d_out, n) # 1 PCIe transfer out
finally:
tacuda.device_free(d_in)
tacuda.device_free(d_out)
device_upload takes its element count from the array. An explicit count may
transfer a prefix but never more than the array holds: the native call receives
only a pointer, so this is checked before it. The raw tacuda._lib entry points
do not have this protection.
Single-series functions require a 1D array and raise ValueError otherwise,
because a matrix would be read as one concatenated series crossing symbol
boundaries. Use the *_batch functions for (numSeries, seriesLen) matrices.
Available Indicators (full TA-Lib coverage)
| Category | Indicators |
|---|---|
| Moving Averages | SMA, EMA, WMA, DEMA, TEMA, TRIMA, KAMA, T3, MAMA, MAVP, MA |
| Momentum | RSI, MACD, MACDEXT, MACDFIX, Stochastic, StochF, StochRSI, CCI, CMO, ROC, ROCP, ROCR, ROCR100, MOM, WILLR, APO, PPO, PVO, TRIX, ULTOSC, BOP |
| Volatility | ATR, NATR, TRANGE, BBANDS, ACCBANDS, StdDev, VAR |
| Trend | ADX, ADXR, DX, +DI, -DI, +DM, -DM, SAR, SAREXT, Aroon, AroonOsc |
| Volume | AD, ADOSC, OBV, MFI, IMI, NVI, PVI |
| Statistical | LINEARREG, SLOPE, INTERCEPT, ANGLE, TSF, CORREL, BETA, AVGDEV, SUM, MAX, MIN |
| Price Transform | AVGPRICE, MEDPRICE, TYPPRICE, WCLPRICE, MIDPRICE |
| Hilbert Transform | HT_DCPERIOD, HT_DCPHASE, HT_PHASOR, HT_SINE, HT_TRENDLINE, HT_TRENDMODE |
| Vector Math (19) | SIN, COS, TAN, ASIN, ACOS, ATAN, SINH, COSH, TANH, EXP, LN, LOG10, SQRT, CEIL, FLOOR, ADD, SUB, MULT, DIV |
| Candlestick | 63 patterns (Doji, Hammer, Engulfing, MorningStar, etc.) |
| Device API | ct_*_d variants of the above + device_alloc/upload/download/sync/free and buffer-pool controls |
Library Location
The bindings search for the native library in this order:
TACUDA_LIBRARYenvironment variable (full path to shared library)TACUDA_LIBRARY_PATH/TACUDA_LIBRARY_DIR+ library name- Native library bundled beside the Python module
- System library path (
ctypes.util.find_library), then working-directory and source-build locations
The loader registers CUDA_PATH\bin while loading a Windows DLL.
An invalid TACUDA_LIBRARY raises a load error. Start a new Python process
after switching DLLs.
Regenerating Metadata
When the C header changes, regenerate from the repository root:
python bindings/generate_bindings.py
Build & Publish (PyPI)
cd bindings/python
python -m build
python -m twine check dist/*
python -m twine upload dist/*
Changelog
Canonical release history is maintained in CHANGELOG.md.
License
Apache License 2.0 — see LICENSE.
Fused portfolio batches (0.9.0)
Seventeen indicator families can process a portfolio in one CUDA kernel
per call. Supported device batches are CUDA Graph compatible; user-stream
calls stay asynchronous. TACUDA_BATCH_FUSED=auto|0|1 controls automatic,
per-series or forced fused dispatch. Period/shape limits apply, and the
double internal arithmetic can change the last float32 bits from 0.8.
See batch dispatch and migration.
Metadata
Release files for tacuda 0.9.3
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| tacuda-0.9.3.tar.gz | 46.7 kB | Without provenance | Details for tacuda-0.9.3.tar.gz |
Built distributions (wheels)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| tacuda-0.9.3-py3-none-win_amd64.whl | Python 3 | none | Windows x86-64 | Details for tacuda-0.9.3-py3-none-win_amd64.whl |
| tacuda-0.9.3-py3-none-manylinux_2_35_x86_64.whl | Python 3 | none | Linux glibc 2.35+ x86-64 | Details for tacuda-0.9.3-py3-none-manylinux_2_35_x86_64.whl |
Total release size: 91.0 MB
Release files / tacuda-0.9.3.tar.gz
| Download URL | tacuda-0.9.3.tar.gz |
|---|---|
| Size | 46.7 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
8cbe3da5593b93f49d0cb6c6631e7d93b93c2e0db997f8fe4b04a931176fcb2e
|
|
BLAKE2b-256 checksum How to use checksums |
4b3f61d4a5db693298c8de9877a03386035a3d270c5277f7f92ef7c92146cce4
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.16
|
Release files / tacuda-0.9.3-py3-none-win_amd64.whl
| Download URL | tacuda-0.9.3-py3-none-win_amd64.whl |
|---|---|
| Size | 45.6 MB |
| Tags | Python 3 Windows x86-64 |
|
SHA-256 checksum How to use checksums |
abcd2bf89c5b8435f8baa40819862949a652718782b79accad6d3708ccc53d35
|
|
BLAKE2b-256 checksum How to use checksums |
e47e4c2daf728d8a6ee0aa28ed097cbd49adc2cbb01949bded7e2ac1d0435b4d
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.16
|
Release files / tacuda-0.9.3-py3-none-manylinux_2_35_x86_64.whl
| Download URL | tacuda-0.9.3-py3-none-manylinux_2_35_x86_64.whl |
|---|---|
| Size | 45.4 MB |
| Tags | Linux glibc 2.35+ x86-64 Python 3 |
|
SHA-256 checksum How to use checksums |
0c0b9109f0c46a36ff85a57e832dba44b005e6940a0fe2296a651a141eb12e05
|
|
BLAKE2b-256 checksum How to use checksums |
4e33b6219434ff7dc8b760bf3dae470c442ee9629c9ae44354a18881484bb7ae
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.16
|
Security for tacuda 0.9.3
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.