A deep neural network with configurable number of hidden layers that classifies cat vs non-cat Images.
An L layer network is created, with RELU activations in all layers except for the output layer that uses a SIGMOID.
Regularization, momentum and mini-batch techniques are NOT used.
Implemented using NumPy.
Execute cat_image_classifier.py, by configuring layers_dims array you can set the number of layers and their size.
By setting layers_dims = [12288, 20, 15, 10, 5, 1]
We create a 4 hidden layer network, with 12288 nodes on input layer, 20 nodes on layer one and so on.
The code will:
- Train the defined model for configurable number of iterations
- Use trained parameters to compute accuracy for train and test datasets
- Try to predict the label on a custom image provided by the user.
Inspiration and help from Coursera deeplearning specialization