eileen-code4fun

import tensorflow as tf

(train_images, train_labels), (test_images, test_labels) = tf.keras.dataset.cifar10.load_data()

def preprocess(filename, images, labels):
  with tf.io.TFRecordWriter(filename) as writer:
    for image, label in zip(images, labels):
      # Encode the image and label in tf.train.Example.
      feature = {
        # Normalize the image to range [0, 1].

eileen-code4fun

import matplotlib.pyplot as plt

class_names = ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck']

dataset = tf.data.TFRecordDataset([GCS_PATH_FOR_DATA + 'train.tfrecord'])
plt.figure(figsize=(10, 10))
for i, example in enumerate(dataset.take(16)):
  data = tf.train.Example()
  data.ParseFromString(example.numpy())
  image = tf.constant(data.features.feature['image'].float_list.value, shape=[32, 32, 3])

eileen-code4fun

def extract(example):
  data = tf.io.parse_example(
    example,
    # Schema of the example.
    {
      'image': tf.io.FixedLenFeature(shape=(32, 32, 3), dtype=tf.float32),
      'label': tf.io.FixedLenFeature(shape=(), dtype=tf.int64)
    }
  )
  return data['image'], data['label']

eileen-code4fun

from tensorflow.keras import layers, models, losses

def create_model():
  model = models.Sequential([
    layers.Conv2D(32, (3, 3), activation='relu', input_shape=(32, 32, 3)),
    layers.MaxPooling2D(2, 2),
    layers.Conv2D(64, (3, 3), activation='relu'),
    layers.MaxPooling2D(2, 2),
    layers.Conv2D(64, (3, 3), activation='relu'),
    layers.Flatten(),

eileen-code4fun

model.fit(train_dataset, epochs=5, validation_data=val_dataset)
model.evaluate(test_dataset, verbose=2)
# Output
# loss: 1.0985 - accuracy: 0.6060

eileen-code4fun

# A distributed strategy to take advantage of available hardward.
# No-op otherwise.
mirrored_strategy = tf.distribute.MirroredStrategy()
with mirrored_strategy.scope():
  model = create_model()
  # Restore from the latest checkpoint if available.
  latest_ckpt = tf.train.latest_checkpoint(GCS_PATH_FOR_CHECKPOINTS)
  if latest_ckpt:
    model.load_weights(latest_ckpt)

eileen-code4fun

import argparse

parser = argparse.ArgumentParser()
parser.add_argument('--epochs', dest='epochs', type=int, default=5)

args = parser.parse_args()

# Refer to the argument as args.epochs.

eileen-code4fun

from setuptools import find_packages
from setuptools import setup

setup(
  name='trainer',
  version='0.1',
  packages=find_packages(),
  include_package_data=True,
  description='An E2E Tutorial of Vertex AI'
)

eileen-code4fun

cloud_model = tf.keras.models.load_model(GCS_PATH_FOR_SAVED_MODEL)
# Compile the model to make sure it uses the correct accuracy metric.
cloud_model.compile(optimizer='adam', loss=tf.keras.losses.SparseCategoricalCrossentropy(), metrics=['accuracy'])
cloud_model.evaluate(test_dataset, verbose=2)
# Output loss: 0.917 - accuracy: 0.6620

eileen-code4fun

workerPoolSpecs:
  machineSpec:
    # Machines and GPUs: https://cloud.google.com/vertex-ai/docs/training/configure-compute#specifying_gpus
    machineType: n1-standard-4
    acceleratorType: NVIDIA_TESLA_V100
    acceleratorCount: 2
  replicaCount: 1
  pythonPackageSpec:
    # Executors: https://cloud.google.com/vertex-ai/docs/training/pre-built-containers
    executorImageUri: us-docker.pkg.dev/vertex-ai/training/tf-gpu.2-3:latest