duots 0.1.0

Composable timeseries feature extraction for paired signals.

duots

duots is a lightweight Python package for calculating features from paired time series signals — like those collected from symmetrical body parts (e.g., left and right wrists). It provides a composable, lazy, and efficient pipeline to build complex signal analysis routines.

---

Features

• Composable feature pipelines using functional programming
• Modular primitives: segmentation, transformation, timeseries ops, value aggregation
• Efficient via functools.lru_cache (minimizes redundant computation)
• Minimal dependencies: uses only scipy (and optionally sampen-gpu)
• Designed for paired signals (e.g., (left, right) or (x, y))
• Easy to extend, debug, and test

Design Philosophy

• Composable: Build powerful feature extractors from simple, small functions.
• Efficient: Shared operations are cached; performance scales with reuse.
• Minimal: Only scipy is used for FFT, skew, and kurtosis; avoids heavy dependencies.

📦 Installation

From PyPI

pip install duots

From Source

git clone https://github.com/4d30/duots.git
cd duots
pip install .

Example Usage

from duots import generate, compose

# Create a pair of signals, e.g., (left, right)
# They must be tuples, of the same length, without NaNs
sig_a = tuple(range(1, 100))
sig_b = tuple(range(1, 100))
signal_pair = (sig_a, sig_b)

# Calculate values for each process
for proc in generate.processes():
    names, funcs = zip(*proc)
    name = compose.descriptors(names)
    composed_function = compose.functions(funcs)
    value = composed_function(signal_pair)
    print(f"{name}: {value}")