A simple neural network implementation which utilizes a simimilar API to TensorFlow's Sequential models.
An in-depth paragraph about your project and overview of use.
- NumPy
pip3 install -r requirements.txt
Below is an example of a neural network with two Dense() layers using Tanh() activation functions learning the XOR function. Although seemingly trivial, the XOR function isn't linearly separable, meaning linear models such as logistic regression and single-layer perceptrons cannot learn XOR.
from nn import Sequential, Input, Dense, Tanh
np.random.seed(42)
# XOR input/output data
X = np.reshape([[0, 0], [0, 1], [1, 0], [1, 1]], (4, 2, 1))
Y = np.reshape([[0], [1], [1], [0]], (4, 1, 1))
# Model instantiation
model = Sequential([
Dense(2, 3),
Tanh(),
Dense(3, 1),
Tanh(),
])
model.compile()
# Model training
model.fit(X, Y, epochs=10000, learning_rate=0.01)
# Predict
Y_pred = nn.predict(X)
for (y_true, y_pred) in zip(Y, Y_pred):
print(f'Actual: {y_true}, Predicted: {y_pred}')Output:
Actual: [[0]], Predicted: [[0.0003956]]
Actual: [[1]], Predicted: [[0.97018558]]
Actual: [[1]], Predicted: [[0.97092169]]
Actual: [[0]], Predicted: [[0.00186825]]
nn.Sequential(
layers=[]
)| Function | Description |
|---|---|
__init__ |
Instantiates a new Sequential model. If given a list of layers, it will add these to the model, similar to add(). |
add |
Add a single layer to the model. |
compile |
Takes the added layers and the parameters of compile to generate a trainable model. |
fit |
After compilation, fit() trains the model on its inputs and outputs. |
predict |
Makes predictions after fitting to the data. Takes in a subset of the input data to make a prediction. |
Abstract class for the layers API. This shouldn't be used in an instance of a model.
This should always be the first layer of the Sequential model. Since the other layers take in an explicit output_size as their input, they infer their input_size from the previous layer's output_size. This means we must declare the model's first input_size.
Just your regular densely-connected NN layer. At the moment, the dense layer only performs the dot product and bias addition, but no activation function. The activation function is out-sourced to the Activation layers.
Applies an activation function to an output.
Hyperbolic tangent activation function.