Jupyter TensorFlow Examples

Mnist Example

(adapted from tensorflow/tensorflow - mnist_softmax.py)

1. When creating your notebook server choose a container image which has Jupyter and TensorFlow installed.

2. Use Jupyter’s interface to create a new Python 3 notebook.

3. Copy the following code and paste it into your notebook:

   from tensorflow.examples.tutorials.mnist import input_data
   mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
   
   import tensorflow as tf
   
   x = tf.placeholder(tf.float32, [None, 784])
   
   W = tf.Variable(tf.zeros([784, 10]))
   b = tf.Variable(tf.zeros([10]))
   
   y = tf.nn.softmax(tf.matmul(x, W) + b)
   
   y_ = tf.placeholder(tf.float32, [None, 10])
   cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y), reduction_indices=[1]))
   
   train_step = tf.train.GradientDescentOptimizer(0.05).minimize(cross_entropy)
   
   sess = tf.InteractiveSession()
   tf.global_variables_initializer().run()
   
   for _ in range(1000):
     batch_xs, batch_ys = mnist.train.next_batch(100)
     sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})
   
   correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
   accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
   print("Accuracy: ", sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels}))
   

4. Run the code. You should see a number of WARNING messages from TensorFlow, followed by a line showing a training accuracy something like this:

   Accuracy:  0.9012
   

Next steps

• See a simple example of creating Kubeflow pipelines in a Jupyter notebook.
• Build machine-learning pipelines with the Kubeflow Pipelines SDK.
• Learn the advanced features available from a Kubeflow notebook, such as submitting Kubernetes resources or building Docker images.