_images/logo1.png

Welcome to PyEPO’s documentation!¶

This is the documentation of PyEPO (PyTorch-based End-to-End Predict-then-Optimize Tool), which aims to provide end-to-end methods for predict-then-optimize tasks.

Quick Example¶

End-to-end training of a shortest-path predictor on a 5x5 grid with the SPO+ loss:

import pyepo
import torch
from torch import nn
from torch.utils.data import DataLoader

# optimization model: 5x5 grid shortest path
grid = (5, 5)
optmodel = pyepo.model.grb.shortestPathModel(grid)

# synthetic data and dataset
x, c = pyepo.data.shortestpath.genData(
    num_data=1000, num_features=5, grid=grid, deg=4, noise_width=0.5, seed=135,
)
dataset = pyepo.data.dataset.optDataset(optmodel, x, c)
dataloader = DataLoader(dataset, batch_size=32, shuffle=True)

# linear predictor and SPO+ loss
predmodel = nn.Linear(5, 40)
spo = pyepo.func.SPOPlus(optmodel, processes=1)
optimizer = torch.optim.Adam(predmodel.parameters(), lr=1e-3)

# end-to-end training
for epoch in range(10):
    for xb, cb, wb, zb in dataloader:
        loss = spo(predmodel(xb), cb, wb, zb)
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

# decision quality
print("Regret:", pyepo.metric.regret(predmodel, optmodel, dataloader))

Contents:

Indices and Tables¶