CODEWITHSUNDEEP
pythontensorflowneural-networksigmoid
Alex Nikitin
Alex Nikitin

Reputation: 524

Sigmoid operation implementation in Tensorflow

For learning purposes I have a task to make a linear and sigmoid operations in tensorflow. I managed to do the linear op:

def linear_op_forward(X, W):
''' linear operation'''
return np.dot(X, W.T)

def linear_op_backward(op, grads):
    ''' Linear gradient realization '''
    X = op.inputs[0]  
    W = op.inputs[1]  
    dX = tf.multiply(grads, W)
    dW = tf.reduce_sum(tf.multiply(X, grads),
                       axis = 0,
                       keep_dims = True)
    return dX, dW

But I'm stuck with sigmoid operation:

Is that correct?

def sigmoid_op_forward(X):
    return 1 / (1 + np.exp(-X))

And I have hard time understandind sigmoid gradient

def sigmoid_op_backward(op, grads):
    ???

Can someone please help with this?

Upvotes: 2

Views: 551

Answers (1)

Maxim
Maxim

Reputation: 53758

Try this:

def sigmoid_op_backward(op, grads):
  sigmoid = op.outputs[0]
  return sigmoid * (1 - sigmoid) * grads

Upvotes: 2

Related Questions