Training a neural network on MNIST with Keras

This simple example demonstrates how to plug TensorFlow Datasets (TFDS) into a Keras model.

View on TensorFlow.org Run in Google Colab View source on GitHub Download notebook 
import tensorflow as tf
import tensorflow_datasets as tfds

Step 1: Create your input pipeline

Start by building an efficient input pipeline using advices from:

Load a dataset

Load the MNIST dataset with the following arguments:

  • shuffle_files=True: The MNIST data is only stored in a single file, but for larger datasets with multiple files on disk, it's good practice to shuffle them when training.
  • as_supervised=True: Returns a tuple (img, label) instead of a dictionary {'image': img, 'label': label}.
(ds_train, ds_test), ds_info = tfds.load(
    'mnist',
    split=['train', 'test'],
    shuffle_files=True,
    as_supervised=True,
    with_info=True,
)

2026-01-15 12:12:40.447201: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected

Build a training pipeline

Apply the following transformations:

def normalize_img(image, label):
  """Normalizes images: `uint8` -> `float32`."""
  return tf.cast(image, tf.float32) / 255., label

ds_train = ds_train.map(
    normalize_img, num_parallel_calls=tf.data.AUTOTUNE)
ds_train = ds_train.cache()
ds_train = ds_train.shuffle(ds_info.splits['train'].num_examples)
ds_train = ds_train.batch(128)
ds_train = ds_train.prefetch(tf.data.AUTOTUNE)

Build an evaluation pipeline

Your testing pipeline is similar to the training pipeline with small differences:

  • You don't need to call tf.data.Dataset.shuffle.
  • Caching is done after batching because batches can be the same between epochs.
ds_test = ds_test.map(
    normalize_img, num_parallel_calls=tf.data.AUTOTUNE)
ds_test = ds_test.batch(128)
ds_test = ds_test.cache()
ds_test = ds_test.prefetch(tf.data.AUTOTUNE)

Step 2: Create and train the model

Plug the TFDS input pipeline into a simple Keras model, compile the model, and train it.

model = tf.keras.models.Sequential([
  tf.keras.layers.Flatten(input_shape=(28, 28)),
  tf.keras.layers.Dense(128, activation='relu'),
  tf.keras.layers.Dense(10)
])
model.compile(
    optimizer=tf.keras.optimizers.Adam(0.001),
    loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
    metrics=[tf.keras.metrics.SparseCategoricalAccuracy()],
)

model.fit(
    ds_train,
    epochs=6,
    validation_data=ds_test,
)

Epoch 1/6
/tmpfs/src/tf_docs_env/lib/python3.9/site-packages/keras/src/layers/reshaping/flatten.py:37: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.
  super().__init__(**kwargs)
469/469 ━━━━━━━━━━━━━━━━━━━━ 4s 4ms/step - loss: 0.6206 - sparse_categorical_accuracy: 0.8293 - val_loss: 0.1876 - val_sparse_categorical_accuracy: 0.9457
Epoch 2/6
469/469 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - loss: 0.1740 - sparse_categorical_accuracy: 0.9514 - val_loss: 0.1374 - val_sparse_categorical_accuracy: 0.9614
Epoch 3/6
469/469 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - loss: 0.1212 - sparse_categorical_accuracy: 0.9656 - val_loss: 0.1098 - val_sparse_categorical_accuracy: 0.9668
Epoch 4/6
469/469 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - loss: 0.0906 - sparse_categorical_accuracy: 0.9724 - val_loss: 0.0974 - val_sparse_categorical_accuracy: 0.9702
Epoch 5/6
469/469 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - loss: 0.0740 - sparse_categorical_accuracy: 0.9792 - val_loss: 0.0894 - val_sparse_categorical_accuracy: 0.9726
Epoch 6/6
469/469 ━━━━━━━━━━━━━━━━━━━━ 2s 3ms/step - loss: 0.0615 - sparse_categorical_accuracy: 0.9822 - val_loss: 0.0858 - val_sparse_categorical_accuracy: 0.9738
<keras.src.callbacks.history.History at 0x7f1d38ba57f0>