This small project is for my education purpose, which I try to implement in Python the main concepts of a vanilla Multilayer Perceptron.
The concepts worked here were:
- Feedfoward process
- Backpropagation
- Gradient Checking
- Create distinct neural net structures
- Use distinct batch sizes
The scripts allow:
- To work with two problems:
linear-regression
andlogistic-regression
; - To instance a neural net with any number of hidden layers and any number of units;
- To use or not the bias elemnt in the layer;
- To use gradient checking;
- To set any size of batch;
The restrictions are:
- Only work with an output of one dimension;
- All activation functions are
softmax function
;
To see an example of how to train a neural net, check the main notebook.