Getting started

Quickstart

Train a small network on XOR and read its predictions.

XOR is the classic problem a single neuron cannot solve, and a small network can.

1. Describe the network

python

from pymlkit import Network, Activations, Losses

model = Network(2)                          # two inputs
model.add_dense(8, Activations.ReLU)        # hidden layer
model.add_dense(1, Activations.Sigmoid)     # output layer, last
model.set_loss(Losses.HalfMSE)

The output layer is whichever one you add last. There is no separate output declaration.

2. Prepare flat data

Samples are concatenated into one flat list. Four samples with two inputs each become eight numbers.

python

x = [0, 0,   0, 1,   1, 0,   1, 1]
y = [0,      1,      1,      0]

3. Fit

python

model.fit(x * 50, y * 50, lr=0.5, epochs=300, batch=4, log_step=100)

Repeating the four samples gives the optimizer more steps per epoch. Training prints progress and a loss value every log_step epochs.

4. Predict

python

for i in range(4):
    sample = x[i * 2 : i * 2 + 2]
    print(sample, "->", round(model.forward(sample)[0], 2))

# [0, 0] -> 0.02
# [0, 1] -> 0.99
# [1, 0] -> 0.99
# [1, 1] -> 0.01

Exact numbers vary between runs because weights are initialised randomly.

Installation

Network