pypolymix 0.1.0

A light-weight PyTorch companion for building stochastic surrogate models

Pypolymix

pypolymix is a light-weight PyTorch companion for building stochastic surrogate models.

Documentation

Project documentation lives at https://sandialabs.github.io/pypolymix.

Installation

From PyPI:

python -m pip install pypolymix

Optional extras:

python -m pip install "pypolymix[examples]"
python -m pip install "pypolymix[docs]"
python -m pip install "pypolymix[dev]"

For local development:

python -m pip install -e ".[dev,docs,examples]"

Quickstart

Fit a 1D polynomial chaos model with a mix of stochastic and deterministic coefficients:

import torch

from pypolymix.parameter_groups import DeterministicGroup, IIDGaussianGroup
from pypolymix.surrogate_models import PolynomialChaosExpansion
from pypolymix import StochasticModel

torch.manual_seed(0)
x = torch.linspace(-1, 1, 200).unsqueeze(-1)
y = torch.sin(3 * x) + 0.1 * torch.randn_like(x)

surrogate_model = PolynomialChaosExpansion(num_inputs=1, num_outputs=1, degree=5)
num_params = surrogate_model.num_params()

parameter_groups = [
    IIDGaussianGroup("stochastic", 2),
    DeterministicGroup("deterministic", num_params - 2),
]

model = StochasticModel(surrogate_model=surrogate_model, parameter_groups=parameter_groups)

optimizer = torch.optim.Adam(model.parameters(), lr=5e-3)
for _ in range(500):
    optimizer.zero_grad()
    preds = model(x, num_samples=16).mean(dim=0)
    data_loss = torch.mean((preds - y) ** 2)
    loss = data_loss + 1e-3 * model.distribution_loss()
    loss.backward()
    optimizer.step()

with torch.no_grad():
    samples = model(x, num_samples=100)
    mean = samples.mean(dim=0)
    std = samples.std(dim=0)

Development

If you use Poetry:

poetry install --with dev,docs,examples

Install the pre-commit hook:

pre-commit install

Examples

Notebooks live in docs/examples/ and are rendered on the documentation site.