adfluo's Introduction

adfluo

adfluo, adfluis, adfluere, adfluxi, adfluxum

to flow on/to/towards/by
to glide/drift quietly

adfluo is a Python library for pipeline-oriented feature computation, mainly aimed at tricky multimodal datasets that might require a wide range of different features to be computed from.

Adfluo makes your feature extraction code:

clean : it encourages you to outline clearly the steps needed to compute a feature as a pipeline of atomic steps
data scientist-friendly: adfluo's output has a predictable structure, ensuring that once you've run the feature extraction, you'll be able to focus a 100% on your data-science/statistics work.
efficient : if different features have computation steps in common, adfluo will do its best to compute only what is necessary, without any extra configuration.
reusable: By separating the input data from the feature computation logic, you'll easily be able to reuse an existing extraction pipeline on another dataset, or use another extraction pipeline on the same dataset.
sample-oriented: adfluo organizes its processing around samples of data.

Installation

Adfluo is available on Pypi, and has no special dependencies, a simple

pip install adfluo

will do.

Example

import random

# Defining our dataset as a list of dicts
my_dataset = [
    {"numbers" : [random.randint(1, 20) for j in range(50)],
     "idx": i}
  for i in range(20)
]

# TODO: examples 
#  - mean, std dev of numbers
#  - "relative" mean using idx