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