main about multitask-learning HOT 1 CLOSED

import random

......

for epoch in range(epochs)
train()

......

train ():
N1 = size(cifar)
N2 = size(tiny)
N3 = size(vgg)
N = (N1+N2+N3)/batch_size
batch_indx = [i for i in range(N)]
random.shuffle(batch_indx)

for indx in batch_indx:
if indx >= N1+N2:
indx = indx-N1-N2
data_name = 'vgg'
elif indx >= N1:
indx = indx-N1
data_name = 'tiny'
else:
data_name = 'cifar'

  inputs, targets = trainloader[data_name][indx]
  ......

from multitask-learning.