deep_learn package contains numpy of implementation of a standard feed forward neural network. This package will be used as teaching material for my upcoming deep learning course. The model class deep_learn.nn.ann which implements feed forward neural network takes in network architecture as object field. The class method fit which trains the model takes train and test data and neural network hyperparameters as arguments. Here every mathematical operation is implemented using numpy to speed up computation. Other than numpy and built-in packages no other specialized libraries are used to build package.
- deep_learn - contains the package which contains the class module of feed forward neural network
- analysis - contains jupyter notebooks that shows example demonstrations of deep learning tasks using the deep_learn package, it also contains the iris_classification notebook which trains neural network to classify iris data (check the notebook)
Download the package. Open the terminal, and change current directory to the directory containing the setup.py file. Install the package using the following command:
pip install .Note: the jypyter notebooks and test.py file can run without pip installation of the package, since these files at the beginning append the path of the package to environment system path if the package is not pip installed.
To create an object of the ann model write the following code:
# import the package
from deep_learn.nn import ann
# size of the layers including the input and output
layers_dims = [4,4,8,8,4,3]
# create model with given architecture
model = ann(layers_dims=layers_dims)To fit the ann model write the following code:
model.fit(X_train, Y_train, X_test, Y_test, batch_size,
learning_rate = learning_rate,
num_iterations = num_iterations, print_cost=True, random_seed = 0)Read the jupyter notebooks in the analysis folder to further understand the usage of the package