Difficulty: Beginner
Estimated Time: 10 minutes

This scenario explains the classification of handwritten digits using TensorFlow.

We will learn the steps, how to download the dataset, visualize image, train a classifier, evaluate, visualize the weights, etc.

The environment has a Docker installation configured, running on a host called docker. Everything else required we'll launch as containers and we will access the Jupyter Notebook in the browser and complete the tutorial.

In this tutorial on Image classification, we learn:

How docker images runs tensorflow and its dependencies environment container. How to download the dataset and visualize images . We have train a classifier, evaluate it and use it to make prediction on new images. Finally we visualize the weights the classifier learns to gain an intuition of how it works. So, digest on working classification of handwritten digits with TF.learn.

Don’t stop now! The next scenario will only take about 10 minutes to complete.

Classification of Handwritten Digits

Step 1 of 5

Step 1

Introduction and Installation of TensorFlow

As Docker is already installed and configured, we are good to go directly downloading and running the tensorflow image.

Before moving further, we'll simply understand the problem is "classifying the handwritten digits from the MNIST dataset and writing a simple classifier for this is often considered as writing a Hello World of computer vision. MNIST is a multi-class classification problem. Given an image of a digit, our job is to predict which one it is. We have written an ipython Notebook(official tutorial from tensorflow community) for this tutorial and to make it easier to configure your environment, we'll start with quick screencast with installing TensorFlow using Docker.

Here are the steps to do it, 1) Installation 2) Download Data set 3) Visualize Images 4) Train a Classifier 5) Evaluate 6) Visualize Weights

Installation of TensorFlow using Docker

You can find the installation instruction for TensorFlow using docker installation on the homepage under the getting stated.

Next we will launch a docker container with a tensorflow image. The image is hosted on docker hub, simply click on the tensorflow image in previous sentence. The image contains tensorflow with all of its dependencies properly configured. Everything is configured with in this image, as docker is configured on this platform. Here, is the command to run this image, with the version you want, as I have choose the latest release, 1.1.0-rc2

docker run -it -p 8888:8888 tensorflow/tensorflow:1.1.0-rc2

If it's a first time you've run the image, it'll be downloaded automatically. And, on subsequent runs it will be cached locally. The image starts automatically and by default, it runs a notebook server. Just note down the token in the link as I shown below from your terminal window.

 Copy/paste this URL into your browser when you connect for the first time,
to login with a token:

Now, you can go to the browser and point it to the IP on port 8888. In this case, simply click on + option besides Terminal and select View HTTP Port 80 on Host 1 and finally enter the port as 8888 in Display a different port It will prompt to the Login into the Jupyter and the above token you can copied into the password and then we'll have an ipython Notebook dashboard that we can experiment in the browser served by the container. Now you can upload the notebook ep7.ipynb through the UI.