adversarials package

Module contents

Adversarial package.

@author

Victor I. Afolabi Artificial Intelligence Expert & Software Engineer. Email: javafolabi@gmail.com | victor.afolabi@zephyrtel.com GitHub: https://github.com/victor-iyiola

@project

File: __init__.py Created on 20 December, 2018 @ 07:20 PM.

@license

MIT License Copyright (c) 2018. Victor I. Afolabi. All rights reserved.

class adversarials.SimpleGAN(size: Union[Tuple[int], int] = 28, channels: int = 1, batch_size: int = 32, **kwargs)

Bases: adversarials.core.base.ModelBase

Simple Generative Adversarial Network.

Methods:

def __init__(self, size: Union[Tuple[int], int]=28, channels: int=1, batch_size: int=32, **kwargs)

def train(self, X_train, epochs: int=10):

def plot_images(self, samples: int=16, step:int=0):

# Plot generated images

Attributes:

G (keras.model.Model): Generator model. D (keras.model.Model): Discriminator model. model (keras.model.Model): Combined G & D model. shape (Tuple[int]): Input image shape.

call(n: int = 1, dim: int = 100)

Inference method. Given a random latent sample. Generate an image.

Args:
samples (int, optional): Defaults to 1. Number of images to

be generated.

dim (int, optional): Defaults to 100. Noise dimension.

Returns:

np.ndarray: Array-like generated images.

property model

Stacked generator-discriminator model.

Returns:

keras.model.Model: Combined G & D model.

plot_images(samples=16, step=0)

Plot and generate images

samples (int, optional): Defaults to 16. Noise samples to generate. step (int, optional): Defaults to 0. Number of training step currently.

property shape

Input image shape.

Returns:

Tuple[int]: image shape.

train(X_train, epochs: int = 10)

Train function to be used after GAN initialization

X_train[np.array]: full set of images to be used