[Guide] How to Install a Full Data Science & Quant Finance Stack on Termux (NumPy, SciPy, Pandas, Statsmodels, Arch)

[ferzhotime]

[Guide] How to Install a Full Data Science & Quant Finance Stack on Termux (NumPy, SciPy, Pandas, Statsmodels, Arch)

Disclaimer & Introduction

Last time when guide was tested and confirmed to work: 2025-07-03

This guide provides a confirmed, step-by-step process to install a complete and complex Python stack for data science and quantitative finance on Termux. The goal is to successfully install arch, a library that depends on notoriously difficult-to-compile packages like numpy, scipy, and statsmodels.

The standard installation methods often fail due to version incompatibilities and the specific compilation environment of Termux on aarch64. This guide is the result of extensive trial and error and provides a reliable path to success by pinning specific package versions and using custom compiler flags.

This is a community contribution. Please follow the steps precisely, as the order and the specific commands are critical.

Requirements

• CPU architecture: aarch64
• Android OS version: 12+ (Tested on Android 15)
• Termux version: 0.118+
• Python version: 3.12.x

Final Installed Package Versions

Following these steps will result in a stable, compatible environment with these key package versions:

• Numpy: 1.26.4 (Crucial for compatibility)
• SciPy: 1.15.0
• Pandas: (Latest version compiled against NumPy 1.26.4)
• Statsmodels: 0.14.2 (Compiled from a specific git tag)
• Arch: (Latest version)

---

Instructions

Do not re-order or modify steps unless you know what you are doing.

Step 1: System Preparation

First, ensure all your installed Termux packages are up-to-date.

yes | pkg upgrade

Step 2: Install All Build Tools & System Dependencies

We will install a comprehensive set of build tools and system libraries in one go. This prevents compilation errors later.

pkg install git python build-essential cmake ninja libopenblas libandroid-execinfo patchelf binutils-is-llvm clang fftw flang

Step 3: Install Python Build Helpers

Next, install the necessary Python tools for building complex packages from source.

pip3 install setuptools wheel packaging pyproject_metadata cython meson-python versioneer setuptools-scm

Step 4: Compile Core Libraries with Version Pinning

This is the most critical part. We will compile numpy and pandas from source and install a specific, compatible version of scipy. This avoids the most common errors.

# Install a compatible version of NumPy (pre-NumPy 2.0)
MATHLIB=m LDFLAGS="-lpython3.12" pip3 install --no-build-isolation --no-cache-dir numpy==1.26.4

# Install Pandas, which will now build against our compatible NumPy
LDFLAGS="-lpython3.12" pip3 install --no-build-isolation --no-cache-dir pandas

# Install a compatible version of SciPy
pip install scipy==1.15.0

Step 5: Compile statsmodels from a Specific Source Version

The latest statsmodels version has compilation issues on Termux. We will clone the source code, check out a known working version (v0.14.2), and compile it with a special flag to fix a common error.

# Clone the repository
git clone https://github.com/statsmodels/statsmodels.git

# Enter the directory
cd statsmodels

# Check out the specific compatible version
git checkout v0.14.2

# Compile and install with the magic CFLAGS flag
CFLAGS="-U__ANDROID_API__ -D__ANDROID_API__=33 -include complex.h" LDFLAGS+="-lpython3.12" python -m pip install . --no-build-isolation

Step 6: Final Installation

With all the difficult dependencies correctly installed, we can now easily install arch and any other packages you need.

# Go back to the home directory
cd

# Install arch
pip install arch

# (Optional) Install other useful libraries, including trading libs
pip install pandas-ta tqdm

Your environment is now ready!

---

Key Insights & Common Tips

Tip

Why Version Pinning? The newest versions of libraries like NumPy (2.x) introduce breaking changes. Pinning to numpy==1.26.4 and scipy==1.15.0 is the key to maintaining compatibility with the rest of the stack.

Important

Why Compile statsmodels? The CFLAGS="-include complex.h" flag is a critical fix that forces the compiler to include a missing header file, resolving a common cpow error on Termux.

Note

Compiler Flags (LDFLAGS, MATHLIB): These are special instructions for the compiler to find the necessary Python and math libraries during the build process, which is often a point of failure in Termux.

Warning

Do not ever run pip install --upgrade pip. The version of pip in Termux is patched by the maintainers. "Upgrading" it will break your environment. Ignore any suggestions to do so.

Credits

Credits for members that helped pave this way: [sylirre](https://github.com/sylirre), [zmvw235](https://github.com/zmvw235)