Basic Usage

This folder contains notebooks for basic usage of the package, e.g. things like training and evaluating models, and saving and loading models. Basically, there are two ways to DMGP:

  1. Load a pre-trained model

$ python examples/bayesian_mnist.py --model [additive_grid_model]
                                    --mode [train_test_mode]
                                    --batch-size [batch_size]
                                    --epochs [epochs]
                                    --lr [learning_rate]
                                    --save_dir [save_directory]
                                    --num_monte_carlo [num_monte_carlo_inference]
                                    --num_mc [num_monte_carlo_training]
                                    --log_dir [logs_directory]

  1. Define your customized model

import torch
import torch.nn as nn
from dmgp.models import DMGP
from dmgp.kernels import LaplaceProductKernel
from dmgp.utils import HyperbolicCrossDesign
from dmgp.layers import AMGP, LinearFlipout

# Define the model
model = DMGP(
    input_dim=784,
    output_dim=10,
    num_layers=3,
    num_inducing=2,
    hidden_dim=8,
    kernel=LaplaceProductKernel(),
    design_class=HyperbolicCrossDesign,
    layer_type=LinearFlipout,
)

# Define the loss
def loss_fn(model, x, y):
    return -model.elbo(x, y)

# Define the optimizer
optimizer = torch.optim.Adam(model.parameters(), lr=1e-3)

# Train the model
for epoch in range(100):
    optimizer.zero_grad()
    loss = loss_fn(model, x, y)
    loss.backward()
    optimizer.step()

# Evaluate the model
model.eval()
y_pred = model(x)
acc = (y_pred.argmax(dim=-1) == y).float().mean()

print(f"Accuracy: {acc}")