Create simple neural network using a lightweight library
Example of usage (create a simple network)
import neura
import numpy as np
# generate a sample (mimic the mnist digits database)
# 28x28 pixel image of a digit
input_shape = (28, 28)
image = np.random.rand(28, 28)
# create the correct output (y_true)
y = np.zeros(10)
y[int(np.random.rand() * 10)] = 1
# create the model architecture
model = neura.model.Model([
neura.layers.Flatten(input_shape=input_shape),
neura.layers.Dense(128, activation=neura.activation.ReLu()),
neura.layers.Dense(10)
], name="MyModel", learning_rate=0.001)
# compile the model with loss function and optimizer
model.compile(
loss=neura.losses.CategoricalCrossEntropy(),
optimizer="adam"
)
# print the model architecture
model.summary()
# try to categorize the image in one of the 10 classes [0, 1, 2, ..., 9]
pred = model.predict(image, verbose=False)
# evaluate the model prediction without training
loss = model.evaluate(image, y)
# print the results
print("\nresults:")
print(pred)
print(y)
print(loss)MyModel
===============================================
Total number of parameters: 102416
===============================================
LAYERS:
0. Flatten nodes: 1 params: 784
1. Dense nodes: 128 params: 100352
2. Dense nodes: 10 params: 1280
===============================================
results:
[ 55.0807834 8.48622231 7.46788936 6.80623906 -7.329365
-120.89559978 41.64848654 -64.82835674 29.99597943 -49.55296296]
[0. 0. 0. 0. 0. 0. 1. 0. 0. 0.]
[9.992007221626415e-17]
- clone this repository
git clone https://github.com/roysmanfo/neura
- install using pip
pip install .