CODEWITHSUNDEEP
tensorflow
vadamsky
vadamsky

Reputation: 48

How to calculate loss in tensorflow?

I've simple model like this:

n_input = 14 
n_out   = 1 

weights = {
    'out': tf.Variable(tf.random_normal([n_input, n_out]))
}
biases = {
    'out': tf.Variable(tf.random_normal([n_out]))
}
def perceptron(input_tensor, weights, biases):
    out_layer_multiplication = tf.matmul(input_tensor, weights['out'])
    out_layer_addition = out_layer_multiplication + biases['out']
    return out_layer_addition

input_tensor = rows
model = perceptron

"rows" dimension is (N, 14) and "out" dimension is (N), where "out" is result of running model with "rows" as "input_tensor".

And I want to calculate loss in tensorflow. Algorythm of calculating is:

ls = 0
for i in range(len(out)-1):
    if out[i] < out[i+1]:
        ls += 1

Where "ls" is model loss. How can I calculate it in tensorflow notation?

Upvotes: 1

Views: 1866

Answers (1)

Giuseppe Marra
Giuseppe Marra

Reputation: 1104

You can do something like this:

l = out.get_shape()[0]
a = out[0:l-1]
b = out[1:l]
c = tf.where(a<b, tf.ones_like(a), tf.zeros_like(a))
return tf.reduce_sum(c)

In practice, a contains out[i] and b contains out[i+1]. c has 1s every time out[i]<out[i+1]. So summing them is equal to make a +1 each time.

Upvotes: 1

Related Questions