This is an updated version of a post from when artificial neural networks (with ojAlgo) were first introduced. Since then new features have been added, the API changed a bit, performance improved and various changes to (other parts of) ojAlgo enable different ways to do things.
The MNIST database is a large image database of handwritten digits that is commonly used for training and testing in the field of machine learning / image categorisation. Information regarding this dataset and various results achieved is widely published.
A correctly modelled/trained neural network should be able to achieve at least 5% error rate on this dataset. Most/all published results are better than that. The largest, most advanced, models have managed 0,35%. That’s almost unbelievably good. ojAlgo currently doesn’t have all the features required to build that kind of model. The model in the program listed below gets about 2% error rate. Here are some sample digits/images from the data set.
The program below (with its dependency on ojAlgo) can do the following:
- Read/parse the files containing the image data and labels.
- Generate the actual images so that you can inspect them. The example images above are generated with that code.
- Print images to console (to sanity check results, as in the console output below)
- Model and train feedforward neural networks:
- Any number of layers
- Any number of input/output nodes per layer
- 5 different activator functions
- 2 different error/loss functions
- Droptouts and L1/L2 regularisation
- Batch training and invoking
- Choose between
The main benefit of using ojAlgo is how easy it is to do this and get good results. Download the example code below (and ojAlgo) and run it. Then start modifying the network structure, learning rate and other things. (You also need to download the data files, and update the various paths in the programs.)
class ReworkedTrainingANN ojAlgo 2021-08-07 Image 0: 7 <=> 7 X++ XXXXXXXXXXXXXXX + ++XXXXXXXXXX+ XX+ XX XX +XX XX +XX XX +XX XX XX+ XXX XX XX+ +XX XXX +XXX +XX Image 1: 2 <=> 2 ++XXX++ +XXXXXXX +XXXX+XXX+ XXX XX+ XX +XX XXX +XXX XXX +XX+ XXX+ XXX XXX+ XXX XXX+ +XXX XXX XXX ++++ XXXXXXXX+++XXXXXXX+ XXXXXXXXXXXXXXX+++ +++++XXX+++ Image 2: 1 <=> 1 X+ +X +X X+ X XX XX +XX +X XX +X+ XX XX +X+ +X+ XX XX +XX XX+ XX Image 3: 0 <=> 0 +XX XXX+ XXXX+ +XXXXX++ XXXXXXXXX XXXXXXXXXXX XXXXX+ +XX+ XXXXX+ XXX+ XXXX +XXX XXX XXX XX XXX+ +XX +XXX XXX XXXX XXX +XXXX XXX XXXXXX XXX XXXXXXXX +XXXXXXXXXXX+ XXXXXXXXXX +XXXXXX+ XXX Image 4: 4 <=> 4 X +X ++ XX +X X+ +X XX XX +X XX XX +XX +X+ XX XX +XX X+ +X+ XX XX+ +XX++++++XXX+ XXXXXXXXXXX +XX +XX +XX +XX +XX XX+ X Image 5: 1 <=> 1 X+ XXX XX+ XXX XXX +XXX XXX XXX +XX+ +XX XXX XX+ +XX XXX +XX+ +XX XXX XXX +X+ ++ Image 6: 4 <=> 4 X+ XX+ XX XX+ +X+ XX XX +X+ XX+ +XX +X+ +XX +XX +XXXX +XXX+ XXXXXXXXXXXX +XX+++ XX+ XX +X +XX +X+ XX +XX XX+ XX++X+ XXXX+ X+ Image 7: 9 <=> 9 +X +XXX XXXX+ +XXXXXX+ XXXXXXXX XX +XXXX +X XXXX XX+ XXX+ +XX ++XXXX XXXXXX XX+ XXX+ +XX ++ XX+ +XX +X+ XX +XX XX X+ XX X Image 8: 5 <=> 5 ++ +XXXXXXXX +XXXXXXXXXXX XXXXXXXXXXXX + XXX++ XX XX+ +XX +XX +XX XXX+ XXXXX++++ + +XXXXXXXXXXXX+ XXXXXXXXXXXX+ XXXXXXXXX +XX+ +XXX XXXXXXX+ +XXXXXX XXXX+ ++ Image 9: 9 <=> 9 +XX+ +XXXXXXXX+ +XXXX++XXXXX+ +XXXX+ X XXXX +XXX+ X XXX XXX +XXXX+ XXX++XX+XXXXXXX XXXXXXXXXXXXXX +XX++XXXXX+ +XXX+ +XXX+ +XXXX XXXX +XXX XXX+ XXXX XXX XXX+ XXX +X+ ========================================================= Error rate: 0.0189