pygpr is a python package to perform basic Gaussian Process (GP) regression tasks, such as sampling function values from a GP, or evaluating the posterior predictive distribution given some data.
See :ref:detailed-description for links to the python classes.
A short example using a squared exponential kernel. You can copy the code blocs below and paste them into a Ipython terminal using the magic function %paste::
import numpy as np
import pylab as plt
plt.ion()
import pygpr
# Create an instance of a squared exponential kernel.
# Hyper parameters are: 1.0 for the amplitude and 0.1 for the length scale.
sek = pygpr.kernels.SquaredExponentialKernel([1.0, 0.1])
# Define the sample input points where the GP will be defined.
x = np.linspace(0.0, 1.0, 400)
# Build a GP instance passing the Kernel instance and the input array.
mygp = pygpr.GaussianProcess(sek, x)
# Sample the GP function values
f = mygp.sample(size=5)
# We can now plot the five samples.
plt.plot(x, f.T)
Interesting things start to happen when we add some data to the story::
# Create a data array with two (noiseless) datapoints.
data = np.array([[0.2, 0.65], [-1.5, 1.]])
mygp.data = data
# Given these data, we can find the
# GP prediction for the function values.
mean, cov = mygp.prediction()
# And we can obtain samples from the predicted functions
fp = mygp.prediction_sample(size=20)
fig = plt.figure()
plt.plot(x, fp.T)
# Add the datapoints
plt.plot(data[0], data[1], 'or', ms=10)
We can add a further datapoint and see how the prediction changes::
data2 = np.array([[0.2, 0.5, 0.65], [-1.5, -1.2, 1.]])
mean, cov = mygp.prediction(data=data2)
fp = mygp.prediction_sample(size=20)
fig = plt.figure()
plt.plot(x, fp.T)
# Add the datapoints
plt.plot(mygp.data[0], mygp.data[1], 'or', ms=10)
A more detailed example will be included in the :ref:tutorial.
.. _detailed-description:
The package implements the GaussianProcess :ref:py:class:gp: GaussianProcess class and the
Contents:
.. toctree:: :maxdepth: 3
rstfiles/installation rstfiles/tutorial rstfiles/api
- :ref:
genindex - :ref:
modindex - :ref:
search