poseigen-binmeths 0.1.3

Binning and split-generation methods for the Poseigen package family

Binmeths

This package implements methods explored in the Binning Methods paper. poseigen_binmeths is the binning and split-generation package in the Poseigen family. It provides utilities for bin construction, bin statistics, balanced sampling, and train/validation/test split strategies used across Poseigen workflows.

Features

poseigen_binmeths is organized into three modules:

• bin_utils: bin parsing and assignment, histogram utilities, objective binning helpers, bin counts and densities, and weighted bin selection.
• split_utils: split generation utilities including random, window, stratified, and akin-based scoring/split selection methods.
• harpoon: balanced per-bin samplers including NumPy and PyTorch implementations, plus overlapping-bin sampling helpers.

Find Epoch Sampling poseigen_trident.

Binning methods are used in the following case studies:

• Synthetic Right-skewed Data
• CTCF-DNA Interactions
• YY1-DNA Interactions
• Enhancer Activity

Installation

Install from PyPI:

pip install poseigen_binmeths

For local development, install from source using your preferred editable-install workflow.

Usage

Import from the package namespace (bin utilities are re-exported):

import poseigen_binmeths as pbm

Or import modules directly:

import poseigen_binmeths.bin_utils as bu
import poseigen_binmeths.split_utils as su
import poseigen_binmeths.harpoon as harp

Example: histogram binning and a random split.

import numpy as np
import poseigen_binmeths.bin_utils as bu
import poseigen_binmeths.split_utils as su

x = np.random.randn(1000)
bind = bu.HistBinning(x, bins=20)
train_idx, test_idx = su.SplitGen_Random(len(bind), proportions=[0.8, 0.2])

Project Status

poseigen_binmeths is in active development and is intended to support binning and sampling workflows across the Poseigen ecosystem.

Related Projects

• poseigen_seaside: shared utilities and metrics foundation.
• poseigen_trident: model training workflows, including epoch sampling integrations.

License

This project is released under the MIT License.