Simple implementation of MLP neural network in NumPy with supporting examples
This a fully functional feedforward neural network library. The implemented features are:
There are two demos to demonstrate capabilities of the library:
There are lots of comments in the code explaining the details
To install required dependencies: make install
.
To run the demo: python3 iris_demo.py
.
The demo is based on the iris dataset, the dataset can be found in dataset/iris/
. It consists of 150 entries.
The accuracy obtained on the validation set is: 98.7%.
In the demo data is shuffled split into train and validation datasets, the netowork is trained and the performance is displayed as a confusion matrix:
To run the demo: python3 digits_mnist_demo.py
.
This demo is based on mnist digits dataset, the dataset can be found in dataset/digits_mnist/
. It consists of 60000 train images and 10000 test images.
The accuracy obtained on the test set is: 97.8%.
Example network architecture:
Example confusion matrix: