https://github.com/ryancswallace/npdb

Parallel NumPy-like interface for large n-dimensional arrays on disk.
https://github.com/ryancswallace/npdb

mapreduce-designpatterns numpy numpy-arrays parallel-computing python

Last synced: about 1 month ago
JSON representation

Parallel NumPy-like interface for large n-dimensional arrays on disk.

• Host: GitHub
• URL: https://github.com/ryancswallace/npdb
• Owner: ryancswallace
• License: mit
• Created: 2017-12-02T18:12:08.000Z (about 7 years ago)
• Default Branch: master
• Last Pushed: 2017-12-10T21:13:09.000Z (about 7 years ago)
• Last Synced: 2024-11-13T17:55:15.741Z (3 months ago)
• Topics: mapreduce-designpatterns, numpy, numpy-arrays, parallel-computing, python
• Language: Python
• Homepage:
• Size: 75.2 KB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# npdb: Large Parallelized NumPy Arrays

[![PyPI version](https://badge.fury.io/py/npdb.svg)](https://badge.fury.io/py/npdb)

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)

[![Build Status](https://travis-ci.org/ryancwallace/npdb.svg?branch=master)](https://travis-ci.org/ryancwallace/npdb)

[![codecov](https://codecov.io/gh/ryancwallace/npdb/branch/master/graph/badge.svg)](https://codecov.io/gh/ryancwallace/npdb)

[![Documentation Status](https://readthedocs.org/projects/npdb/badge/?version=latest)](http://npdb.readthedocs.io/en/latest/?badge=latest)

[**Docs**](https://npdb.readthedocs.io)

| [**Install Guide**](https://npdb.readthedocs.io/en/latest/install.html)

| [**Tutorial**](https://npdb.readthedocs.io/en/latest/tutorial.html)

| [**Examples**](https://github.com/ryancwallace/npdb/tree/master/algs)

npdb is an implementation of large disk-stored NumPy-compatible n-dimenstional arrays that may exceed available memory. npdb implements the core multi-dimensional array class `npdb.dbarray`, which supports persistent binary storage and distributed batch processed operations. `npdb.dbarray` supports a subset of the `numpy.ndarray` interface.

## Background

The `numpy.memmap` class also supports arrays stored on disk but has several limitations. Arrays are stored in a single file, and the size of the file can not exceed 2GB on 32-bit systems. This implementation both restricts the size of data and disallows distributed storage. On the other hand, `numpy.memmap` objects support the entire `numpy.ndarray` interface. 

The npdb library strikes a different balance--array sizes are constrained only by available disk space and can be distributed across multiple files. The cost of this capability is that a limited subset of the numpy interface is supported.

## Example

```python

import npdb as nd

# create on disk a 3D array of floats of lengths 100

db_arr = npdb.dbarray((100,100,100), float)

# slice from array

```

## Installation

You can install npdb using pip by running

```

$ pip install npdb

```

## Testing

If unittest is installed, tests can be run after installation with

```

$ python -m unittest discover

```

## License

MIT License (see `LICENSE`). Copyright (c) 2017 Ryan Wallace.

## Authors

Ryan Wallace. [email protected].

# IN PRE-ALHPA DEV

TODO: 

---infrastruture---

* autodoc for reference 

* host

* link examples, install guide, docs, tutorial

---core content---

* set minimum size before spill over onto disk

* better handling of open(file) re exceptions,  remove file on delete

* magic methods 

* indexing location, module names?

* creation: basic - npdb.empty(), from data - npdb.array()

* free dbarray? parameter to save explicitly, or paramater for persistence? overload del

* keep track of what's in memory to avoid repetitive pulls

* default size params, data dir

* rewrite arraymap and indexing in Cython/C/numpy?

* compression