Untrained Full Example

This full example creates the an XOR model, trains the model and saves it.

from EasyNN.model import Network, ReLU, LogSoftMax

import EasyNN.callbacks as cb
import numpy as np

# Setup the XOR dataset.
data = np.array([
    [0,0],
    [0,1],
    [1,0],
    [1,1]
] * 100, dtype=int)

labels = np.array([0,1,1,0] * 100, dtype=int)

# Create the Neural Network Model.
model = Network(
    16, ReLU,
    2, LogSoftMax,
)

# Set your models data
model.training.data = (data, labels)

# Set the models labels
model.labels = {
    0: 0,
    1: 1
}

# Change to a small learning rate.
model.optimizer.lr = 0.01

 
# Set when to terminate point. Training will end once your
# validation accuracy hits above 90% two times.
model.callback(
    cb.ReachValidationAccuracy(limit=0.90, patience=2),
)

# Print the accuracy and iteration count every 10 iterations.
model.print.on_validation_start(iteration=True,accuracy=True)
model.print.on_training_start(iteration=True, frequency=10)

# Always at the end of your setup
model.train()

# Save your model so that you can use it later.
model.save("xor")

Output:

Iteration: 0
Iteration: 10
Iteration: 20
Iteration: 30
Iteration: 40
Iteration: 44, Validation Accuracy: 0.76171875
Iteration: 50
Iteration: 60
Iteration: 70
Iteration: 80
Iteration: 89, Validation Accuracy: 1.0
Iteration: 90
Iteration: 100
Iteration: 110
Iteration: 120
Iteration: 130
Iteration: 135, Validation Accuracy: 1.0
Parameters saved.

Loading Trained/Saved model example:

This example is using the XOR model example above.

from EasyNN.model import Model
import numpy as np

# Load the trained model we saved.
model = Model.load("xor")

# Classify an example.
print(model.classify(np.array([0,1])))

Output:

1