Building a Multi-Layer-Perceptron using Tensorflow

Overview

The MNIST database of handwritten digits, available from this page, has a training set of 60,000 examples, and a test set of 10,000 examples. It is a subset of a larger set available from NIST. The digits have been size-normalized and centered in a fixed-size image.

Here we will be building a multi-layer perceptron to classify handwritten digits using tensorflow.

Directions

It is recommended to install tensorflow as a virtual environment via conda.

Install tensorflow
Download the dataset
Split it into training and testing datasets
Test the model
Calculate the accuracy

Dependencies

get the input_data py script and put it on your current working folder.
tensorflow

Purpose

A Deep Learning exercise to classify hand-written characters.

Results Example

run the main.py file:

$ python main.py
  Extracting ~/DS_ML/Deep Learning/train-images-idx3-ubyte.gz
  Extracting ~/DS_ML/Deep Learning/train-labels-idx1-ubyte.gz
  Extracting ~/DS_ML/Deep Learning/t10k-images-idx3-ubyte.gz
  Extracting ~/DS_ML/Deep Learning/t10k-labels-idx1-ubyte.gz
                    
  Iteration: 0001 cost= 29.860465115
  Iteration: 0003 cost= 21.000626442
  Iteration: 0005 cost= 20.136222584
  Iteration: 0007 cost= 19.652899717
  Iteration: 0009 cost= 19.383099742
  Iteration: 0011 cost= 19.165024177
  Iteration: 0013 cost= 18.770430475
  Iteration: 0015 cost= 18.776208125
  Iteration: 0017 cost= 18.655625496
  Iteration: 0019 cost= 18.493998495
  Iteration: 0021 cost= 18.481863666
  Iteration: 0023 cost= 18.312887948
  Iteration: 0025 cost= 18.326373666
  Iteration: 0027 cost= 18.227951255
  Iteration: 0029 cost= 18.035528241
  
  Accuracy: 0.9235
  Tuning completed!

Credits

Credit to google for creating tensorflow. Thanks to Yann LeCun, Courant Institute, NYU; Corinna Cortes, Google Labs, New York; and Christopher J.C. Burges, Microsoft Research, Redmond; All are part of the MNIST DATABASE.