airontools 0.1.24

Machine learning tools to complement the AIronSuit package.

AIronTools

AIronTools (Beta) is a Python library that provides the user with deep learning tools built to work with tensorflow as a backend.

Key features:

1. Model constructor that allows multiple models to be optimized in parallel across multiple GPUs.
2. Block constructor to build customised blocks/models.
3. Layer constructor to build customised layers.
4. Preprocessing utils.

Installation

pip install airontools

Custom Keras subclass to build a variational autoencoder (VAE) with airontools and compatible with aironsuit

import numpy as np
from numpy.random import normal
from tensorflow.keras.optimizers import Adam

from airontools.constructors.models.unsupervised.vae import VAE


tabular_data = np.concatenate(
    [
        normal(loc=0.5, scale=1, size=(100, 10)),
        normal(loc=-0.5, scale=1, size=(100, 10)),
    ]
)
model = VAE(
        input_shape=tabular_data.shape[1:],
        latent_dim=3,
)
model.compile(optimizer=Adam(learning_rate=0.001))
model.fit(
    tabular_data,
    epochs=10,
)
print("VAE evaluation:", float(model.evaluate(tabular_data)["loss"]))

More examples

see usage examples in aironsuit/examples