Example of handling MNIST as a 784-dimensional vector input in ludwig.ai

mnist_vector.py

import os
import logging
from ludwig.api import LudwigModel

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'

config = {
    'preprocessing': {
        'split': {
            'type': 'fixed',
            'column': 'split'
        }
    },
    'input_features': [
        {
            'name': 'image',
            'type': 'vector',
            'encoder': {
                'type': 'dense',
                'fc_layers': [
                    {'output_size': 64},
                    {'output_size': 32}
                ]
            }
        }
    ],
    'output_features': [
        {'name': 'label', 'type': 'category'}
    ],
    'trainer': {'epochs': 1, 'batch_size': 64, 'eval_batch_size': 128}
}


model = LudwigModel(
    config = config, 
    gpus=0,
    logging_level=logging.INFO
)


# mnist_vector.parquet scheme
#  	                                            image 	label 	split
# 0 	0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ... 	0    	0
# 1 	0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ... 	0   	0
# 2 	0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ... 	0   	0
# 3 	0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ... 	0   	0
# 4 	0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ... 	0   	0
# ...
train_stats, _, _ = model.train(dataset='./mnist_vector.parquet')

model_ecd = model.model
print(model_ecd)