Exploring MNIST with TensorFlow

There are a lot of awesome tutorials about how to classify handwritten digits from MNIST database, so my plan is to put some of these tutorials together, evolving from a very simple model to a nice model that achieves 99% accuracy in the test data.

Here is an android app that I did as a final result of this exercise 😄!

Have fun!

Note: most of the code presented here is originally from the tutorials in the Reference section, please follow those for more details

Getting started

For running the "tutorials" You'll need:

• Python 2 or 3
• Numpy
• TensorFlow
• Jupyter Notebook

Getting the data

The MNIST databased used here was the one available at TensorFlow (read more about it), so you don't have to do much.
You can download the data here, if you want.

For the examples in this repository we'll use all the available data and train the models with stochastic gradient descent (except 01-KNN).

Let's do it!

To follow the tutorial samples, run: jupyter notebook.
If you want to check TensorBoard: run tensorboard --logdir=/tmp/tensorflow_log

References

Thank you!

0. TF and DL without a Phd by Martin Gorner, Google
1. Classifying Handwritten Digits with TF.Learn, by Josh Gordon
2. How to Make a TF Image Classifier, by Siraj Raval
3. TensorFlow Tutorials, by Hvass Labs
4. CS231n: Convolutional Neural Networks for Visual Recognition
5. Getting started with TensorFlow
6. Not another mnist tutorial with tensorflow
7. Training convolutional neural network for image classification