import torch
import numpy as np
import math

target = torch.tensor([0, 1.0, 0])
input = torch.tensor([.333, .333, .334])

print("nats: ", -math.log(.333))
print("bits: ", -math.log(.333, 2))
b = torch.nn.functional.cross_entropy(input, target).item()
print(b)
print(b / math.log(2))