Skip to main content

A framework for composing Neural Processes in Python

Project description

Neural Processes

CI Coverage Status Latest Docs Code style: black

A framework for composing Neural Processes in Python. See also NeuralProcesses.jl.

This package is currently under construction. There will be more here soon. In the meantime, see NeuralProcesses.jl.

Contents:

Installation

See the instructions here. Then simply

pip install neuralprocesses

Examples of Predefined Models

TensorFlow

GNP

import lab as B
import tensorflow as tf

import neuralprocesses.tensorflow as nps

cnp = nps.construct_gnp(dim_x=2, dim_y=3, likelihood="lowrank")
dist = cnp(
    B.randn(tf.float32, 16, 2, 10),
    B.randn(tf.float32, 16, 3, 10),
    B.randn(tf.float32, 16, 2, 15),
)
mean, var = dist.mean, dist.var

print(dist.logpdf(B.randn(tf.float32, 16, 3, 15)))
print(dist.sample())
print(dist.kl(dist))
print(dist.entropy())

ConvGNP

import lab as B
import tensorflow as tf

import neuralprocesses.tensorflow as nps

cnp = nps.construct_convgnp(dim_x=2, dim_y=3, likelihood="lowrank")

dist = cnp(
    B.randn(tf.float32, 16, 2, 10),
    B.randn(tf.float32, 16, 3, 10),
    B.randn(tf.float32, 16, 2, 15),
)
mean, var = dist.mean, dist.var

print(dist.logpdf(B.randn(tf.float32, 16, 3, 15)))
print(dist.sample())
print(dist.kl(dist))
print(dist.entropy())

ConvGNP with Auxiliary Variables

import lab as B
import tensorflow as tf

import neuralprocesses.tensorflow as nps

cnp = nps.construct_convgnp(
    dim_x=2,
    dim_yc=(
        3,  # Observed data has three dimensions.
        1,  # First auxiliary variable has one dimension.
        2,  # Second auxiliary variable has two dimensions.
    ),
    dim_xt_aug=4,  # Third auxiliary variable has four dimensions.
    dim_yt=3,  # Predictions have three dimensions.
    num_basis_functions=64, 
    likelihood="lowrank",
)

observed_data = (
    B.randn(tf.float32, 16, 2, 10),
    B.randn(tf.float32, 16, 3, 10),
)

# Define three auxiliary variables. The first one is specified like the observed data
# at arbitrary inputs.
aux_var1 = (
    B.randn(tf.float32, 16, 2, 12),
    B.randn(tf.float32, 16, 1, 12),  # Has one dimension.
)
# The second one is specified on a grid.
aux_var2 = (
    (B.randn(tf.float32, 16, 1, 25), B.randn(tf.float32, 16, 1, 35)),
    B.randn(tf.float32, 16, 2, 25, 35),  # Has two dimensions.
)
# The third one is specific to the target inputs. We could encoder it like the first
# auxiliary variable `aux_var1`, but we illustrate how an MLP-style encoding can
# also be used. The number must match the number of target inputs!
aux_var3 = B.randn(tf.float32, 16, 4, 15)  # Has four dimensions.

dist = cnp(
    [observed_data, aux_var1, aux_var2],
    (B.randn(tf.float32, 16, 2, 15), aux_var3),
)
mean, var = dist.mean, dist.var

print(dist.logpdf(B.randn(tf.float32, 16, 3, 15)))
print(dist.sample())
print(dist.kl(dist))
print(dist.entropy())

PyTorch

GNP

import lab as B
import torch

import neuralprocesses.torch as nps

cnp = nps.construct_gnp(dim_x=2, dim_y=3, likelihood="lowrank")
dist = cnp(
    B.randn(torch.float32, 16, 2, 10),
    B.randn(torch.float32, 16, 3, 10),
    B.randn(torch.float32, 16, 2, 15),
)
mean, var = dist.mean, dist.var

print(dist.logpdf(B.randn(torch.float32, 16, 3, 15)))
print(dist.sample())
print(dist.kl(dist))
print(dist.entropy())

ConvGNP

import lab as B
import torch

import neuralprocesses.torch as nps

cnp = nps.construct_convgnp(dim_x=2, dim_y=3, likelihood="lowrank")
dist = cnp(
    B.randn(torch.float32, 16, 2, 10),
    B.randn(torch.float32, 16, 3, 10),
    B.randn(torch.float32, 16, 2, 15),
)
mean, var = dist.mean, dist.var

print(dist.logpdf(B.randn(torch.float32, 16, 3, 15)))
print(dist.sample())
print(dist.kl(dist))
print(dist.entropy())

Build Your Own Model

ConvGNP

TensorFlow

import lab as B
import tensorflow as tf

import neuralprocesses.tensorflow as nps

dim_x = 1
dim_y = 1

# CNN architecture:
unet = nps.UNet(
    dim=dim_x,
    in_channels=2 * dim_y,
    out_channels=(2 + 512) * dim_y,
    channels=(8, 16, 16, 32, 32, 64),
)

# Discretisation of the functional embedding:
disc = nps.Discretisation(
    points_per_unit=64,
    multiple=2**unet.num_halving_layers,
    margin=0.1,
    dim=dim_x,
)

# Create the encoder and decoder and construct the model.
encoder = nps.FunctionalCoder(
    disc,
    nps.Chain(
        nps.PrependDensityChannel(),
        nps.SetConv(scale=2 / disc.points_per_unit),
        nps.DivideByFirstChannel(),
    ),
)
decoder = nps.Chain(
    unet,
    nps.SetConv(scale=2 / disc.points_per_unit),
    nps.LowRankGaussianLikelihood(512),
)
convgnp = nps.Model(encoder, decoder)

# Run the model on some random data.
dist = convgnp(
    B.randn(tf.float32, 16, 1, 10),
    B.randn(tf.float32, 16, 1, 10),
    B.randn(tf.float32, 16, 1, 15),
)

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

neuralprocesses-0.1.4.tar.gz (31.3 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page