Tensorflow : Cross Entropy for Multi labels classi

Question

I'm using a neural network for classification of a set of objects into n-classes. Each object can belong to multiple classes at the same time (multi-class, multi-label).

For example :

Input is ['a','b','c']  and labels are  [1,0,0,1,1,1]  # where 1 means it is in class and 0 means it's not in class  it's multi labels classification

I search and found people are suggesting you should use sigmoid instead of softmax in Multi-label problems.

some references are :

Sentiment and genres classification

reddit group suggestion on multi label classification

Excellent blog on explaining the theory of multi label classification

another good tutorial on multi label

Stack exchange question

Now after reading those things , I got it that i should go for sigmoid butI have few confusions :

so for single classification what i was doing :

After model (hiddenlayers with lstm ) i was building fully connected layer :

    #weights
    Wo = tf.get_variable('Wo', 
                    shape=[hdim*2, num_labels], 
                    dtype=tf.float32, 
                    initializer=tf.random_uniform_initializer(-0.01, 0.01))


   #bias
   bo = tf.get_variable('bo', 
                    shape=[num_labels,], 
                    dtype=tf.float32, 
                    initializer=tf.random_uniform_initializer(-0.01, 0.01))


  #logits
  logits = tf.matmul(tf.concat([fsf.c, fsb.c], axis=-1), Wo) + bo


   #probability 
   probs  = tf.nn.softmax(logits)

   #taking max probability
   preds  = tf.argmax(probs, axis=-1)


   # Cross Entropy
   ce = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=logits, labels=labels)

   #loss
   loss = tf.reduce_mean(ce)

   # Accuracy
   accuracy = tf.reduce_mean(
                            tf.cast(
                                tf.equal(tf.cast(preds, tf.int32), labels),
                                tf.float32
                                )
  #training                         )
  trainop = tf.train.AdamOptimizer().minimize(loss)

But In multi-labels how should i take inputs :

so what i tried :

   #weights
    Wo = tf.get_variable('Wo', 
                    shape=[hdim*2, num_labels], 
                    dtype=tf.float32, 
                    initializer=tf.random_uniform_initializer(-0.01, 0.01))


   #bias
   bo = tf.get_variable('bo', 
                    shape=[num_labels,], 
                    dtype=tf.float32, 
                    initializer=tf.random_uniform_initializer(-0.01, 0.01))


  #logits
  logits = tf.matmul(tf.concat([fsf.c, fsb.c], axis=-1), Wo) + bo


   #probability 
   probs  = tf.nn.sigmoid(logits)

   #taking max probability
   preds  = tf.round(probs)




   # Cross Entropy
   ce = tf.nn.sigmoid_cross_entropy_with_logits(logits=logits, labels=labels)

   #loss
   loss = tf.reduce_mean(ce)

   # Accuracy
   accuracy = tf.reduce_mean(
                            tf.cast(
                                tf.equal(tf.cast(preds, tf.int32), labels),
                                tf.float32
                                )
  #training                         )
  trainop = tf.train.AdamOptimizer().minimize(loss)

Is this right approach , AM i doing right or need correction somewhere ?

If i passing to tf.nn.sigmoid should i need to pass to tf.nn.sigmoid_cross_entropy_with_logits or not ??

Thanks You .