jutility 0.0.22

Collection of Python utilities intended to be useful for machine learning research and experiments.

jutility

Collection of Python utilities intended to be useful for machine learning research and experiments.

Contents

• jutility
  • Contents
  • Install with pip
  • Usage examples
    • plotting
      • Making a simple plot
      • Multiple lines in a single Line
      • Vector fields (with optional normalisation)
      • Shared colour bar
    • util
    • sweep
  • Unit tests
  • Build package locally
  • Updating package on PyPI

Install with pip

The jutility package is available as a Python package on PyPI, and can be installed with pip using the following commands:

python -m pip install -U pip
python -m pip install -U jutility

Alternatively, jutility can be installed in "editable mode" from the GitHub repository:

git clone https://github.com/jakelevi1996/jutility.git
cd jutility
python -m pip install -U pip
python -m pip install -e .

Usage examples

plotting

Making a simple plot

import numpy as np
from jutility import plotting

x = np.linspace(-10, 10, 100, dtype=np.float32)
plotting.plot(
    plotting.Line(x, np.log(1 + np.exp(x)), label="Softplus", c="b"),
    plotting.Line(x, np.maximum(x, 0),      label="ReLU",     c="r", ls="--"),
    legend=True,
    plot_name="Softplus vs ReLU",
    dir_name="images",
)

import numpy as np
from jutility import plotting

rng = np.random.default_rng(0)
x = np.linspace(0, 2)
f = lambda x: x + 0.1 * rng.normal(size=x.shape)

plotting.plot(
    plotting.Line(x, f(x), c="b", marker="o", label="Blue data"),
    plotting.Line(x, np.exp(-f(x)), c="r", marker="o", label="Red data"),
    axis_properties=plotting.AxisProperties("x label", "y label"),
    legend=True,
    plot_name="Simple plot",
    dir_name="images",
)

Multiple lines in a single Line

import numpy as np
from jutility import plotting, util

n = 100
batch_size = 20
rng = np.random.default_rng(0)
noise = rng.normal(size=[n, batch_size])

x = np.linspace(-1, 4, n).reshape(n, 1)
y = np.exp(-x) + 0.1 * noise

plotting.plot(
    plotting.Line(x, y, alpha=0.2,       c="b", zorder=10),
    plotting.Line(x, np.mean(y, axis=1), c="r", zorder=20),
    plot_name="Multiple lines in a single Line",
    dir_name="images",
)

Vector fields (with optional normalisation)

import numpy as np
from jutility import plotting, util

n = 25
x = np.linspace(-2, 2, n).reshape(1, n)
y = np.linspace(-2, 2, n).reshape(n, 1)
dx = y - x
dy = x + y
mp = plotting.MultiPlot(
    plotting.Subplot(
        plotting.Quiver(x, y, dx, dy, zorder=10, normalise=False),
        axis_properties=plotting.AxisProperties(title="normalise=False")
    ),
    plotting.Subplot(
        plotting.Quiver(x, y, dx, dy, zorder=10, normalise=True),
        axis_properties=plotting.AxisProperties(title="normalise=True")
    ),
    figure_properties=plotting.FigureProperties(figsize=[10, 4])
)
mp.save(plot_name="Vector field", dir_name="images")

Shared colour bar

import numpy as np
from jutility import plotting

rng = np.random.default_rng(0)
z1 = rng.random((100, 200)) + 5
z2 = rng.random((100, 200)) + 2
v_min = min(z1.min(), z2.min())
v_max = max(z1.max(), z2.max())

colour_bar = plotting.ColourBar(v_min, v_max)

mp = plotting.MultiPlot(
    plotting.ImShow(c=z1, vmin=v_min, vmax=v_max),
    colour_bar,
    plotting.ImShow(c=z2, vmin=v_min, vmax=v_max),
    colour_bar,
    figure_properties=plotting.FigureProperties(
        num_rows=2,
        num_cols=2,
        width_ratios=[1, 0.2],
        tight_layout=False,
        title="Shared colour bar",
    ),
)
mp.save("Shared colour bar", dir_name="images")

More complex examples coming soon

util

Coming soon

sweep

Coming soon

(in the meantime, see scripts/make_logo.py which made the logo above, and unit tests for util, plotting, and sweep)

Unit tests

To run all unit tests, install pytest (these tests have previously been run with pytest version 5.4.1), and run the following command (at the time of writing, this takes about 17 seconds to run 42 unit tests, because several unit tests involve saving images or GIFs to disk):

pytest

Build package locally

jutility can be installed in "editable mode" using pip using the following command:

python -m pip install -e .

Alternatively, jutility can be built and installed locally using the following commands, replacing $WHEEL_NAME with the name of the wheel built by the python -m build command (for example, jutility-0.0.5-py3-none-any.whl):

python -m build
python -m pip install --force-reinstall --no-deps dist/$WHEEL_NAME

Or alternatively, run the script build_local.py, which automatically extracts the version number to infer the name of the wheel and then builds it:

python build_local.py

Updating package on PyPI

This package was uploaded to PyPI following the Packaging Python Projects tutorial in the official Python documentation.

To update PyPI with a newer version, update the version tag in setup.cfg, and then use the following commands:

rm -rf dist/*
python -m build
python -m twine upload dist/*

When prompted by twine, enter __token__ as the username, and paste an API token from the PyPI account management webpage as the password (including the pypi- prefix).