¡Google I/O es una envoltura! Póngase al día con las sesiones de TensorFlow Ver sesiones

Cargar datos NumPy

Ver en TensorFlow.org Ejecutar en Google Colab Ver fuente en GitHub Descargar libreta

Este tutorial proporciona un ejemplo de cómo cargar datos de matrices NumPy en un tf.data.Dataset .

Este ejemplo carga el conjunto de datos MNIST desde un archivo .npz . Sin embargo, la fuente de las matrices NumPy no es importante.

Configuración

import numpy as np
import tensorflow as tf

Cargar desde archivo .npz

DATA_URL = 'https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz'

path = tf.keras.utils.get_file('mnist.npz', DATA_URL)
with np.load(path) as data:
  train_examples = data['x_train']
  train_labels = data['y_train']
  test_examples = data['x_test']
  test_labels = data['y_test']

Cargue matrices NumPy con tf.data.Dataset

Suponiendo que tiene una matriz de ejemplos y una matriz de etiquetas correspondiente, pase las dos matrices como una tupla a tf.data.Dataset.from_tensor_slices para crear un tf.data.Dataset .

train_dataset = tf.data.Dataset.from_tensor_slices((train_examples, train_labels))
test_dataset = tf.data.Dataset.from_tensor_slices((test_examples, test_labels))

Usar los conjuntos de datos

Mezclar y procesar por lotes los conjuntos de datos

BATCH_SIZE = 64
SHUFFLE_BUFFER_SIZE = 100

train_dataset = train_dataset.shuffle(SHUFFLE_BUFFER_SIZE).batch(BATCH_SIZE)
test_dataset = test_dataset.batch(BATCH_SIZE)

Construir y entrenar un modelo

model = tf.keras.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.RMSprop(),
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['sparse_categorical_accuracy'])
model.fit(train_dataset, epochs=10)
Epoch 1/10
938/938 [==============================] - 3s 2ms/step - loss: 3.5318 - sparse_categorical_accuracy: 0.8762
Epoch 2/10
938/938 [==============================] - 2s 2ms/step - loss: 0.5408 - sparse_categorical_accuracy: 0.9289
Epoch 3/10
938/938 [==============================] - 2s 2ms/step - loss: 0.3770 - sparse_categorical_accuracy: 0.9473
Epoch 4/10
938/938 [==============================] - 2s 2ms/step - loss: 0.3281 - sparse_categorical_accuracy: 0.9566
Epoch 5/10
938/938 [==============================] - 2s 2ms/step - loss: 0.2940 - sparse_categorical_accuracy: 0.9621
Epoch 6/10
938/938 [==============================] - 2s 2ms/step - loss: 0.2622 - sparse_categorical_accuracy: 0.9657
Epoch 7/10
938/938 [==============================] - 2s 2ms/step - loss: 0.2446 - sparse_categorical_accuracy: 0.9698
Epoch 8/10
938/938 [==============================] - 2s 2ms/step - loss: 0.2147 - sparse_categorical_accuracy: 0.9739
Epoch 9/10
938/938 [==============================] - 2s 2ms/step - loss: 0.1956 - sparse_categorical_accuracy: 0.9750
Epoch 10/10
938/938 [==============================] - 2s 2ms/step - loss: 0.1964 - sparse_categorical_accuracy: 0.9759
<keras.callbacks.History at 0x7fc7a80beb50>
model.evaluate(test_dataset)
157/157 [==============================] - 0s 2ms/step - loss: 0.7089 - sparse_categorical_accuracy: 0.9572
[0.7088937163352966, 0.9571999907493591]