Here, we use a CNN that is similar to Yann LeCun's LeNet 5 to learn handwritten digit recognition.

Download the MNIST data set from THE MNIST DATABASE of handwritten digits. You need all four files. Create the directory "mnist" in your working directory, move the data set to this directory and execute the benchmark or pass the directory of the MNIST data set as argument to the program. Some information about the classification of the test set will be logged in the file "evaluation-*.log", where '*' is the starting time.

To execute the benchmark you can run the Python script:

python benchmark.py [download] [run] [evaluate]

download will download the dataset, run will start the benchmark and evaluate will plot the result. You can of course modify the script or do each step manually.