What is this?

An MNIST classifier!

The final architecture ended up being:

The net was trained over 7 epochs with a learning rate of 0.297635 (optimal lr holding the epoch and hidden layer counts constant).

The final accuracy was 98.44%.

Updates

Keeping the architecture constant, I was able to boost the performance to over 99% (99.1%) by training the net:

Over 10 epochs
With a learning rate of 0.088587 (found to be optimal given the architecture and epoch count)

Name		Name	Last commit message	Last commit date
Latest commit History 12 Commits
.ipynb_checkpoints		.ipynb_checkpoints
mnist_dataset		mnist_dataset
my_own_images		my_own_images
.gitignore		.gitignore
README.md		README.md
constants.py		constants.py
deep_model.json		deep_model.json
deep_net.py		deep_net.py
imagerecreation.ipynb		imagerecreation.ipynb
learning_rate_ex.csv		learning_rate_ex.csv
load_own_images.ipynb		load_own_images.ipynb
m_calc.py		m_calc.py
makeyourownnn.ipynb		makeyourownnn.ipynb
mm.py		mm.py
net.py		net.py
nn.ipynb		nn.ipynb
nn_mnist_backwards.ipynb		nn_mnist_backwards.ipynb
opt_lr.py		opt_lr.py
train_deep.py		train_deep.py