loss = F.cross_entropy(logits, target)
pred = logits.argmax(dim=1, keepdim=True)
LOG_temp['acc'] += [
pred.eq(target.view_as(pred)).sum().item()
/ pred.size(0)
]
LOG_temp['cls_loss'] += [loss.item()]
LOG_temp['gen_loss'] += [
F.mse_loss(x_recon, data).item()
]
current_acc[task_t] = mean_fn(LOG_temp['acc'])
logging_per_task(
wandb, LOG, run, mode, 'acc', task, task_t,
np.round(np.mean(LOG_temp['acc']), 2))
logging_per_task(
wandb, LOG, run, mode, 'cls_loss', task,
task_t,
np.round(np.mean(LOG_temp['cls_loss']), 2))
logging_per_task(
wandb, LOG, run, mode, 'gen_loss', task,
task_t,
np.round(np.mean(LOG_temp['gen_loss']), 2))
print('\n{}:'.format(mode))
print(LOG[run][mode]['acc'])
# store the best accuracy seen so far to all the tasks
best_acc_yet[mode] = np.maximum(
x_mean,
data,
z_mu,
z_var,
z0,
zk,
ldj,
args,
beta=args.beta)
LOG_temp['gen_loss'] += [gen_loss.item()]
# End Minibatch Eval Loop
#-------------------
logging_per_task(wandb, LOG, run, mode, 'acc', task,
task_t,
np.round(np.mean(LOG_temp['acc']), 2))
logging_per_task(
wandb, LOG, run, mode, 'cls_loss', task, task_t,
np.round(np.mean(LOG_temp['cls_loss']), 2))
logging_per_task(
wandb, LOG, run, mode, 'gen_loss', task, task_t,
np.round(np.mean(LOG_temp['gen_loss']), 2))
# End Task Eval Loop
#-------------------
print('\n{}:'.format(mode))
print(LOG[run][mode]['acc'])
# End Eval Loop