losses.py

import math
import torch
import numpy as np
def log_normal_diag(z, mu, logvar):
    return -0.5 * (math.log(2 * math.pi) + logvar + (z - mu).pow(2) / logvar.exp())
def loss_function(recon_mu, recon_logvar, x, z, mu, logvar):
    x = x.view(x.size(0), -1)
    recon_mu = recon_mu.view(x.size(0), -1)
    recon_logvar = recon_logvar.view(x.size(0), -1)
    BCE = -(log_normal_diag(x, recon_mu, recon_logvar)).sum(1).mean()
    log_q = log_normal_diag(z, mu, logvar)
    log_p = log_normal_diag(z, z * 0, z * 0)
    KLD_element = log_q - log_p
    KLD = KLD_element.sum(1).mean()
    return BCE, KLD

def L1_loss_function(recon_mu,x):
    x = x.view(x.size(0), -1)
    recon_mu = recon_mu.view(x.size(0), -1)
    return torch.abs(recon_mu-x).sum(1).mean()
    #return torch.abs(recon_mu-x).mean()

def euclidean(x, y):
    x = x.cpu().detach().numpy()
    y = y.cpu().detach().numpy()
    error = np.zeros(x.shape[0])
    dist = np.zeros(x.shape[0])
    for i, sample in enumerate(x):
        recon = y[i]
        errors = np.zeros(sample.shape[0])
        dists = np.zeros(sample.shape[0])
        for j, v in enumerate(sample):
            dists[j] = np.sqrt(np.power(v - recon[j], 2).sum())
            if dists[j] > 0.001:
                errors[j] = 1
        error[i] = errors.sum() / sample.shape[0]
        dist[i] = dists.sum() / sample.shape[0]
    return error.sum() / x.shape[0], dist.sum() / x.shape[0]