Recomendar películas: recuperación con estrategia de distribución

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

En este tutorial, vamos a entrenar el mismo modelo de recuperación como lo hicimos en la recuperación básica tutorial, pero con la estrategia de distribución.

Iremos a:

  1. Obtenga nuestros datos y divídalos en un conjunto de prueba y entrenamiento.
  2. Configure dos GPU virtuales y TensorFlow MirroredStrategy.
  3. Implemente un modelo de recuperación con MirroredStrategy.
  4. Ajústelo con MirrorredStrategy y evalúelo.

Importaciones

Primero saquemos nuestras importaciones del camino.

pip install -q tensorflow-recommenders
pip install -q --upgrade tensorflow-datasets
import os
import pprint
import tempfile

from typing import Dict, Text

import numpy as np
import tensorflow as tf
import tensorflow_datasets as tfds
import tensorflow_recommenders as tfrs

Preparando el conjunto de datos

Preparamos el conjunto de datos exactamente de la misma manera como lo hacemos en la recuperación básica tutorial.

# Ratings data.
ratings = tfds.load("movielens/100k-ratings", split="train")
# Features of all the available movies.
movies = tfds.load("movielens/100k-movies", split="train")

for x in ratings.take(1).as_numpy_iterator():
  pprint.pprint(x)

for x in movies.take(1).as_numpy_iterator():
  pprint.pprint(x)

ratings = ratings.map(lambda x: {
    "movie_title": x["movie_title"],
    "user_id": x["user_id"],
})
movies = movies.map(lambda x: x["movie_title"])

tf.random.set_seed(42)
shuffled = ratings.shuffle(100_000, seed=42, reshuffle_each_iteration=False)

train = shuffled.take(80_000)
test = shuffled.skip(80_000).take(20_000)

movie_titles = movies.batch(1_000)
user_ids = ratings.batch(1_000_000).map(lambda x: x["user_id"])

unique_movie_titles = np.unique(np.concatenate(list(movie_titles)))
unique_user_ids = np.unique(np.concatenate(list(user_ids)))

unique_movie_titles[:10]
{'bucketized_user_age': 45.0,
 'movie_genres': array([7]),
 'movie_id': b'357',
 'movie_title': b"One Flew Over the Cuckoo's Nest (1975)",
 'raw_user_age': 46.0,
 'timestamp': 879024327,
 'user_gender': True,
 'user_id': b'138',
 'user_occupation_label': 4,
 'user_occupation_text': b'doctor',
 'user_rating': 4.0,
 'user_zip_code': b'53211'}
2021-10-14 11:16:44.748468: W tensorflow/core/kernels/data/cache_dataset_ops.cc:768] The calling iterator did not fully read the dataset being cached. In order to avoid unexpected truncation of the dataset, the partially cached contents of the dataset  will be discarded. This can happen if you have an input pipeline similar to `dataset.cache().take(k).repeat()`. You should use `dataset.take(k).cache().repeat()` instead.
{'movie_genres': array([4]),
 'movie_id': b'1681',
 'movie_title': b'You So Crazy (1994)'}
2021-10-14 11:16:45.396856: W tensorflow/core/kernels/data/cache_dataset_ops.cc:768] The calling iterator did not fully read the dataset being cached. In order to avoid unexpected truncation of the dataset, the partially cached contents of the dataset  will be discarded. This can happen if you have an input pipeline similar to `dataset.cache().take(k).repeat()`. You should use `dataset.take(k).cache().repeat()` instead.
array([b"'Til There Was You (1997)", b'1-900 (1994)',
       b'101 Dalmatians (1996)', b'12 Angry Men (1957)', b'187 (1997)',
       b'2 Days in the Valley (1996)',
       b'20,000 Leagues Under the Sea (1954)',
       b'2001: A Space Odyssey (1968)',
       b'3 Ninjas: High Noon At Mega Mountain (1998)',
       b'39 Steps, The (1935)'], dtype=object)

Configurar dos GPU virtuales

Si no ha agregado aceleradores de GPU a su Colab, desconecte el tiempo de ejecución de Colab y hágalo ahora. Necesitamos la GPU para ejecutar el siguiente código:

gpus = tf.config.list_physical_devices("GPU")
if gpus:
  # Create 2 virtual GPUs with 1GB memory each
  try:
    tf.config.set_logical_device_configuration(
        gpus[0],
        [tf.config.LogicalDeviceConfiguration(memory_limit=1024),
         tf.config.LogicalDeviceConfiguration(memory_limit=1024)])
    logical_gpus = tf.config.list_logical_devices("GPU")
    print(len(gpus), "Physical GPU,", len(logical_gpus), "Logical GPUs")
  except RuntimeError as e:
    # Virtual devices must be set before GPUs have been initialized
    print(e)

strategy = tf.distribute.MirroredStrategy()
Virtual devices cannot be modified after being initialized
INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0',)
INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0',)

Implementando un modelo

Ponemos en práctica las user_model, movie_model, métricas y de trabajo de la misma manera como lo hacemos en la recuperación básica tutorial, pero nos envolvemos en el ámbito estrategia de distribución:

embedding_dimension = 32

with strategy.scope():
  user_model = tf.keras.Sequential([
    tf.keras.layers.StringLookup(
        vocabulary=unique_user_ids, mask_token=None),
    # We add an additional embedding to account for unknown tokens.
    tf.keras.layers.Embedding(len(unique_user_ids) + 1, embedding_dimension)
  ])

  movie_model = tf.keras.Sequential([
    tf.keras.layers.StringLookup(
        vocabulary=unique_movie_titles, mask_token=None),
    tf.keras.layers.Embedding(len(unique_movie_titles) + 1, embedding_dimension)
  ])

  metrics = tfrs.metrics.FactorizedTopK(
    candidates=movies.batch(128).map(movie_model)
  )

  task = tfrs.tasks.Retrieval(
    metrics=metrics
  )
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).
INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).

Ahora podemos ponerlo todo junto en un modelo. Este es exactamente el mismo que en la recuperación básica tutorial.

class MovielensModel(tfrs.Model):

  def __init__(self, user_model, movie_model):
    super().__init__()
    self.movie_model: tf.keras.Model = movie_model
    self.user_model: tf.keras.Model = user_model
    self.task: tf.keras.layers.Layer = task

  def compute_loss(self, features: Dict[Text, tf.Tensor], training=False) -> tf.Tensor:
    # We pick out the user features and pass them into the user model.
    user_embeddings = self.user_model(features["user_id"])
    # And pick out the movie features and pass them into the movie model,
    # getting embeddings back.
    positive_movie_embeddings = self.movie_model(features["movie_title"])

    # The task computes the loss and the metrics.
    return self.task(user_embeddings, positive_movie_embeddings)

Ajuste y evaluación

Ahora instanciamos y compilamos el modelo dentro del alcance de la estrategia de distribución.

Tenga en cuenta que estamos utilizando Adam optimizador aquí en lugar de Adagrad como en la recuperación básica tutorial desde Adagrad no se admite aquí.

with strategy.scope():
  model = MovielensModel(user_model, movie_model)
  model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.1))

Luego, mezcle, agrupe y almacene en caché los datos de capacitación y evaluación.

cached_train = train.shuffle(100_000).batch(8192).cache()
cached_test = test.batch(4096).cache()

Luego entrena el modelo:

model.fit(cached_train, epochs=3)
2021-10-14 11:16:50.692190: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:461] The `assert_cardinality` transformation is currently not handled by the auto-shard rewrite and will be removed.
Epoch 1/3
10/10 [==============================] - 8s 328ms/step - factorized_top_k/top_1_categorical_accuracy: 5.0000e-05 - factorized_top_k/top_5_categorical_accuracy: 8.2500e-04 - factorized_top_k/top_10_categorical_accuracy: 0.0025 - factorized_top_k/top_50_categorical_accuracy: 0.0220 - factorized_top_k/top_100_categorical_accuracy: 0.0537 - loss: 70189.8047 - regularization_loss: 0.0000e+00 - total_loss: 70189.8047
Epoch 2/3
10/10 [==============================] - 3s 329ms/step - factorized_top_k/top_1_categorical_accuracy: 3.3750e-04 - factorized_top_k/top_5_categorical_accuracy: 0.0113 - factorized_top_k/top_10_categorical_accuracy: 0.0251 - factorized_top_k/top_50_categorical_accuracy: 0.1268 - factorized_top_k/top_100_categorical_accuracy: 0.2325 - loss: 66736.4560 - regularization_loss: 0.0000e+00 - total_loss: 66736.4560
Epoch 3/3
10/10 [==============================] - 3s 332ms/step - factorized_top_k/top_1_categorical_accuracy: 0.0012 - factorized_top_k/top_5_categorical_accuracy: 0.0198 - factorized_top_k/top_10_categorical_accuracy: 0.0417 - factorized_top_k/top_50_categorical_accuracy: 0.1834 - factorized_top_k/top_100_categorical_accuracy: 0.3138 - loss: 64871.2997 - regularization_loss: 0.0000e+00 - total_loss: 64871.2997
<keras.callbacks.History at 0x7fb74c479190>

Puede ver en el registro de entrenamiento que TFRS está utilizando ambas GPU virtuales.

Finalmente, podemos evaluar nuestro modelo en el conjunto de prueba:

model.evaluate(cached_test, return_dict=True)
2021-10-14 11:17:05.371963: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:461] The `assert_cardinality` transformation is currently not handled by the auto-shard rewrite and will be removed.
5/5 [==============================] - 4s 193ms/step - factorized_top_k/top_1_categorical_accuracy: 5.0000e-05 - factorized_top_k/top_5_categorical_accuracy: 0.0013 - factorized_top_k/top_10_categorical_accuracy: 0.0043 - factorized_top_k/top_50_categorical_accuracy: 0.0639 - factorized_top_k/top_100_categorical_accuracy: 0.1531 - loss: 32404.8092 - regularization_loss: 0.0000e+00 - total_loss: 32404.8092
{'factorized_top_k/top_1_categorical_accuracy': 4.999999873689376e-05,
 'factorized_top_k/top_5_categorical_accuracy': 0.0013000000035390258,
 'factorized_top_k/top_10_categorical_accuracy': 0.00430000014603138,
 'factorized_top_k/top_50_categorical_accuracy': 0.06385000050067902,
 'factorized_top_k/top_100_categorical_accuracy': 0.1530500054359436,
 'loss': 29363.98046875,
 'regularization_loss': 0,
 'total_loss': 29363.98046875}

Con esto concluye el tutorial de recuperación con estrategia de distribución.