MNIST com qualidade de código

from setuptools import setup, find_packages

setup(
    name='mnist',
    version='1.0.0',
    packages=find_packages(),
    url='',
    license='',
    author='André Claudino',
    author_email='',
    description='',
    install_requires=[
        "tensorflow-gpu==2.4.0",
        "click==7.1.2"
    ]
)

setup.py

main.py

Parseia linha de comando

Configura o ambiente

Carrega dados de treino

Instancia modelo

Treina o modelo

Salva o modelo

from typing import Tuple

import click

import tensorflow as tf

from mnist.models.gray_image_classifier import GrayImageClassifier
from mnist.persistence.microdata import load_grayscale_images
from mnist.training.loop import run_training_loop

main.py

Importa o que for preciso

@click.command()
@click.option("--dataset-path", type=click.STRING, required=True, help="Path for the dataset used for training")
@click.option("--output-path", type=click.STRING, default="output",
              help="path where checkpoints, metrics and model artifact will be saved")
@click.option("--batch-size", type=click.INT, default=32, help="Training batch size")
@click.option("--images-height", type=click.INT, required=True, help="final height of images after resize")
@click.option("--images-width", type=click.INT, required=True, help="Final width of images after resize")
@click.option("--epochs", type=click.INT, default=1, help="Number of training epochs (repeats of dataset)")
@click.option("--learning-rate", type=click.FLOAT, required=True, help="Leargning rate for gradient optimization")
@click.option("--debug/--no-debug", default=False, help="Should or not use tensorflow in debug mode")
@click.option("--layer-sizes", "_layer_size_string", type=click.STRING,
              help="Comma-separeted list of dense layer sizes for the model")
@click.option("--number-of-classes", type=click.INT,
              help="Number of output classes (the number os neurons in the output layer)")
@click.option("--summary-step-size", type=click.INT, default=10,
              help="Number of steps between each metric report and checkpoint save")
def main(dataset_path: str, output_path: str, batch_size: int, images_height: int, images_width: int,
         epochs: int, learning_rate: float, debug: bool, _layer_size_string: Tuple[int], number_of_classes: int,
         summary_step_size: int):

    # Parse layer list
    layer_sizes = _parse_layer_sizes(_layer_size_string)

main.py

Parseia linha de comando