vicolab / ml-pyxis

Tool for reading and writing datasets of tensors in a Lightning Memory-Mapped Database (LMDB). Designed to manage machine learning datasets with fast reading speeds.

ml-pyxis

Tool for reading and writing datasets of tensors (numpy.ndarray) with MessagePack and Lightning Memory-Mapped Database (LMDB).

Example

import numpy as np
import pyxis as px

# Create data
nb_samples = 10
X = np.ones((nb_samples, 2, 2), dtype=np.float32)
y = np.arange(nb_samples, dtype=np.uint8)

# Write
db = px.Writer(dirpath='data', map_size_limit=1)
db.put_samples('input', X, 'target', y)
db.close()

# Read
db = px.Reader(dirpath='data')
sample = db[0]
db.close()

print(sample)

{'input': array([[ 1.,  1.], [ 1.,  1.]], dtype=float32), 'target': array(0, dtype=uint8)}

More examples can be found in the examples/ directory.

Installation

The installation instructions are generic and should work on most operating systems that support the prerequisites.

ml-pyxis requires Python version 2.7, 3.4, 3.5, or 3.6. We recommend installing ml-pyxis, as well as all prerequisites, in a virtual environment via virtualenv.

Prerequisites

The following Python packages are required to use ml-pyxis:

• lmdb - Universal Python binding for the LMDB 'Lightning' Database
• msgpack - MessagePack implementation for Python (binary serialisation)
• NumPy - N-dimensional array object and tools for operating on them
• six - A Python 2 and 3 compatibility library

Please refer to the individual packages for more information about additional dependencies and how to install them for your operating system.

Bleeding-edge installation

To install the latest version of ml-pyxis, use the following command:

pip install --upgrade https://github.com/vicolab/ml-pyxis/archive/master.zip

Add the --user tag if you want to install the package in your home directory.

Notice

The previous LMDB-only API has been deprecated in favour of a combination between LMDB and msgpack. The old version can be installed by using the following commit hash with pip:

pip install --upgrade git+git://github.com/vicolab/ml-pyxis.git@787c3484e3121f2767b254fc41be091d0a3e0cf0

Development installation

ml-pyxis can be installed from source in such a way that any changes to your local copy will take effect without having to reinstall the package. Start by making a copy of the repository:

git clone https://github.com/vicolab/ml-pyxis.git

Next, enter the directory and install ml-pyxis in development mode by issuing the following command:

cd ml-pyxis
python setup.py develop