def __init__(self,
num_samples=3000,
step_size=0.1,
num_steps_per_sample=1,
tau_out=1,
n_units_1=50,
n_units_2=50,
n_units_3=50,
keep_every=2,
normalize_input=True,
normalize_output=True,
seed=0,
RM=False,
burnin=500,
prior_tau=1.0):
"""
This module performs HMC approximation for a Bayesian neural network.
"""
# Set the random seeds
self.seed = seed
torch.manual_seed(seed)
hamiltorch.set_random_seed(seed)
# Check availability of GPU
self.device = torch.device(
"cuda: 0" if torch.cuda.is_available() else "cpu")
self.X = None
self.y = None
# Network params
self.network = None
self.normalize_input = normalize_input
self.normalize_output = normalize_output
self.n_units_1 = n_units_1
self.n_units_2 = n_units_2
self.n_units_3 = n_units_3
# HMC params
self.num_steps_per_sample = num_steps_per_sample
self.step_size = step_size
self.num_samples = num_samples
self.RM = RM
self.burnin = burnin
self.tau_out = tau_out
self.prior_tau = prior_tau
self.keep_every = keep_every
import torch
import hamiltorch
import matplotlib.pyplot as plt
import torch.nn as nn
import numpy as np
import util
# device
print(f'Is CUDA available?: {torch.cuda.is_available()}')
# device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
device = 'cpu'
# hyperparameters
hamiltorch.set_random_seed(123)
prior_std = 1
like_std = 0.1
step_size = 0.001
burn = 100
num_samples = 200
L = 100
layer_sizes = [1, 16, 16, 1]
activation = torch.tanh
pde = True
pinns = False
epochs = 10000
tau_priors = 1 / prior_std**2
tau_likes = 1 / like_std**2
lb = 0