What is Weighted Crossentropy in TF learn explain with example

Recipe Objective

This recipe explains what is Weighted Crossentropy.

Step 1: Importing Libraries

We'll import tflearn, tensorflow as tf and tflearn.datasets.mnist as mnist.

import tflearn
import tensorflow as tf
import tflearn.datasets.mnist as mnist
from __future__ import division, print_function, absolute_import


Step 2: Building Model

We have combined TFLearn built-in ops with Tensorflow graph. We have built this using MNIST Dataset.
To create a multilayer perceptron we have used TFLearn PReLU activations ops.

with tf.Graph().as_default():

    x = tf.placeholder("float", [None, 392])
    y = tf.placeholder("float", [None, 5])

    u = tf.Variable(tf.random_normal([392, 128]))
    v = tf.Variable(tf.random_normal([128, 128]))
    w = tf.Variable(tf.random_normal([128, 5]))
    a = tf.Variable(tf.random_normal([128]))
    b = tf.Variable(tf.random_normal([128]))
    c = tf.Variable(tf.random_normal([5]))

    def net(X):
       X = tflearn.prelu(tf.add(tf.matmul(X, u), a))
       tflearn.summaries.monitor_activation(x)
       X = tflearn.prelu(tf.add(tf.matmul(X, v), b))
       tflearn.summaries.monitor_activation(x)
       X = tf.nn.softmax(tf.add(tf.matmul(X, w), c))


       return X

    my_net = net(x)


Step 3: Weighted Crossentropy

It's syntax is : tflearn.objectives.weighted_crossentropy (y_pred, y_true, weight). Its argument are y_pred which is predicted values, y_true which is labeled targets and weights are a coefficient to use on the positive examples.
It computes weighted sigmoid cross entropy between my_net (logits) and y (labels) and weighted cross entropy.

Loss = tflearn.weighted_crossentropy(my_net, y, weights=0.8)
accuracy = tflearn.metrics.accuracy_op(my_net, y)