Training AmoebaNet-D on Cloud TPU

Objective: This tutorial shows you how to train AmoebaNet-D on Cloud TPU. You can apply the same pattern to other TPU-optimized image classification models that use TensorFlow and the ImageNet dataset.

The AmoebaNet-D model is one of the image classifier architectures discovered using Evolutionary AutoML. The model is based on results from the AmoebaNet paper: Real, E., Aggarwal, A., Huang, Y. and Le, Q.V., 2018, Regularized Evolution for Image Classifier Architecture Search, arXiv preprint arXiv:1802.01548.

This model uses TPUEstimator —a high-level TensorFlow API—which is the recommended way to build and run a machine learning model on a Cloud TPU.

Warning: TPUEstimator is only supported by Tensorflow 1.x. If you are writing a model with Tensorflow 2.x, use Keras instead.

The API simplifies the model development process by hiding most of the low-level implementation, making it easier to switch between TPU and other platforms such as GPU or CPU.

Warning: Amoebanet uses a third-party dataset. Google provides no representation, warranty, or other guarantees about the validity, or any other aspects of this dataset.

Objectives

• Create a Cloud Storage bucket to hold your dataset and model output.
• Prepare a test version of the ImageNet dataset, referred to as the fake_imagenet dataset.
• Run the training job.
• Verify the output results.

Costs

This tutorial uses the following billable components of Google Cloud:

• Compute Engine
• Cloud TPU
• Cloud Storage

To generate a cost estimate based on your projected usage, use the pricing calculator. New Google Cloud users might be eligible for a free trial.

Before you begin

Before starting this tutorial, check that your Google Cloud project is correctly set up.

1. Sign in to your Google Cloud account. If you're new to Google Cloud, create an account to evaluate how our products perform in real-world scenarios. New customers also get $300 in free credits to run, test, and deploy workloads.

2. In the Google Cloud console, on the project selector page, select or create a Google Cloud project.

   Note: If you don't plan to keep the resources that you create in this procedure, create a project instead of selecting an existing project. After you finish these steps, you can delete the project, removing all resources associated with the project.

   Go to project selector

3. Make sure that billing is enabled for your Cloud project. Learn how to check if billing is enabled on a project.

4. In the Google Cloud console, on the project selector page, select or create a Google Cloud project.

   Note: If you don't plan to keep the resources that you create in this procedure, create a project instead of selecting an existing project. After you finish these steps, you can delete the project, removing all resources associated with the project.

   Go to project selector

5. Make sure that billing is enabled for your Cloud project. Learn how to check if billing is enabled on a project.

This walkthrough uses billable components of Google Cloud. Check the Cloud TPU pricing page to estimate your costs. Be sure to clean up resources you create when you've finished with them to avoid unnecessary charges.

Set up your resources

This section provides information on setting up Cloud Storage bucket, VM, and Cloud TPU resources for tutorials.

Important: Set up all resources in the same region/zone to reduce network latency and network costs.

1. Open a Cloud Shell window.

   Open Cloud Shell

2. Create a variable for your project's ID.

   export PROJECT_ID=project-id
   

3. Configure Google Cloud CLI to use the project where you want to create Cloud TPU.

   gcloud config set project ${PROJECT_ID}
   

   The first time you run this command in a new Cloud Shell VM, an Authorize Cloud Shell page is displayed. Click Authorize at the bottom of the page to allow gcloud to make GCP API calls with your credentials.

4. Create a Service Account for the Cloud TPU project.

   gcloud beta services identity create --service tpu.googleapis.com --project $PROJECT_ID
   

   The command returns a Cloud TPU Service Account with following format:

   [e-mail]
   

5. Create a Cloud Storage bucket using the following command:

   gsutil mb -p ${PROJECT_ID} -c standard -l europe-west4 -b on gs://bucket-name
   

   This Cloud Storage bucket stores the data you use to train your model and the training results. The ctpu up tool used in this tutorial sets up default permissions for the Cloud TPU Service Account. If you want finer-grain permissions, review the access level permissions.

   The bucket location must be in the same region as your virtual machine (VM) and your TPU node. VMs and TPU nodes are located in specific zones, which are subdivisions within a region.

6. Launch the Compute Engine and Cloud TPU resources required for this using the ctpu up command.

   Note: This tutorial uses TensorFlow 1.15.5 which is based on the Debian 10 image and defaults to Python 3.6. Earlier versions of TensorFlow 1.15.x were based on Debian 9 and default to Python 3.5.

   ctpu up --project=${PROJECT_ID} \
    --zone=europe-west4-a \
    --vm-only \
    --disk-size-gb=300 \
    --machine-type=n1-standard-8 \
    --tf-version=1.15.5 \
    --name=amoebanet-tutorial

   Command flag descriptions

   project

   Your GCP project ID

   zone

   The zone where you plan to create your Cloud TPU.

   vm-only

   Creates the VM without creating a Cloud TPU. By default the ctpu up command creates a VM and a Cloud TPU.

   disk-size-gb

   The size of the hard disk in GB of the VM created by the ctpu up command.

   machine-type

   The machine type of the Compute Engine VM to create.

   tf-version

   The version of Tensorflow ctpu installs on the VM.

   name

   The name of the Cloud TPU to create.

   For more information on the CTPU utility, see CTPU Reference.

7. When prompted, press y to create your Cloud TPU resources.

   Note: The first time you run ctpu up on a project it takes about 5 minutes to perform startup tasks such as SSH key propagation and API turnup.

When the ctpu up command has finished executing, verify that your shell prompt has changed from username@projectname to username@vm-name. This change shows that you are now logged into your Compute Engine VM. If you are not connected to the Compute Engine instance, you can do so by running the following command:

gcloud compute ssh amoebanet-tutorial --zone=europe-west4-a

From this point on, a prefix of (vm)$ means you should run the command on the Compute Engine VM instance.

Prepare the data

The training application expects your training data to be accessible in Cloud Storage. The training application also uses your Cloud Storage bucket to store checkpoints during training.

Train and evaluate the AmoebaNet-D model with fake_imagenet

ImageNet is an image database. The images in the database are organized into a hierarchy, with each node of the hierarchy depicted by hundreds and thousands of images.

This tutorial uses a demonstration version of the full ImageNet dataset, referred to as fake_imagenet. This demonstration version allows you to test the tutorial, while reducing the storage and time requirements typically associated with running a model against the full ImageNet database.

The fake_imagenet dataset is only useful for understanding how to use a Cloud TPU and validating end-to-end performance. The accuracy numbers and saved model will not be meaningful.

For information on how to download and process the full ImageNet dataset, see Downloading, preprocessing, and uploading the ImageNet dataset.

Caution: For this tutorial, make sure you don't set the STORAGE_BUCKET environment variable to the path of the fake_imagenet dataset. You can read from gs://cloud-tpu-test-datasets but you can't write to it. As a result, you can't use it to write out training logs. Make sure the STORAGE_BUCKET environment variable is set to your own Cloud Storage bucket, as shown above.

In the following steps, a prefix of (vm)$ means you should run the command on your Compute Engine VM.

Note: If you want to monitor the model's output and performance, follow the guide to setting up TensorBoard.

1. Launch a Cloud TPU resource using the ctpu utility.

   (vm)$ ctpu up --project=${PROJECT_ID} \
   --tpu-only \
    --name=amoebanet-tutorial \
    --tf-version=1.15.5
   

2. Set up the following environment variables, replacing bucket-name with the name of your Cloud Storage bucket:

   (vm)$ export STORAGE_BUCKET=gs://bucket-name
   

   (vm)$ export MODEL_DIR=${STORAGE_BUCKET}/amoebanet
   (vm)$ export DATA_DIR=gs://cloud-tpu-test-datasets/fake_imagenet
   (vm)$ export PYTHONPATH="${PYTHONPATH}:/usr/share/tpu/models"
   

   (vm)$ export TPU_NAME=amoebanet-tutorial
   

3. Navigate to the directory:

   (vm)$ cd /usr/share/tpu/models/official/amoeba_net/
   

4. Run the training script.

   (vm)$ python3 amoeba_net.py \
      --tpu=${TPU_NAME} \
      --data_dir=${DATA_DIR} \
      --model_dir=${MODEL_DIR} \
      --num_epochs=1
   

   Parameter	Description
   tpu	Specifies the name of the Cloud TPU. Note that ctpu passes this name to the Compute Engine VM as an environment variable (TPU_NAME).
   data_dir	Specifies the Cloud Storage path for training input. It is set to the fake_imagenet dataset in this example.
   model_dir	Specifies the directory where checkpoints and summaries are stored during model training. If the folder is missing, the program creates one. When using a Cloud TPU, the model_dir must be a Cloud Storage path (`gs://...`). You can reuse an existing folder to load current checkpoint data and to store additional checkpoints as long as the previous checkpoints were created using TPU of the same size and TensorFlow version.

For a single Cloud TPU device, this procedure trains the AmoebaNet-D model for 1 epoch and evaluates every fixed number of steps. Using the specified flags, the model should train in about 30 minutes on a v3-8 Cloud TPU. To train to convergence takes about 90 epochs and approximately 10 hours.

Since the training and evaluation was done on the fake_imagenet dataset, the output results do not reflect actual output that would appear if the training and evaluation was performed on a real dataset.

At this point, you can either conclude this tutorial and clean up your GCP resources, or you can further explore running the model on Cloud TPU Pods.

Scaling your model with Cloud TPU Pods

You can get results faster by scaling your model with Cloud TPU Pods. The fully supported AmoebaNet model can work with the following Pod slices:

• v2-32
• v3-32

When working with Cloud TPU Pods, you first train the model using a Pod, then use a single Cloud TPU device to evaluate the model.

Training with Cloud TPU Pods

Note: If you have already deleted your Compute Engine instance, create a new one following the steps in Set up your resources.

1. Delete the Cloud TPU resource you created for training the model on a single device.

   (vm)$ ctpu delete --project=${PROJECT_ID} \
    --tpu-only \
    --name=amoebanet-tutorial \
    --zone=europe-west4-a

2. Run the ctpu up command, using the tpu-size parameter to specify the Pod slice you want to use. For example, the following command uses a v2-32 Pod slice.

   (vm)$ ctpu up --project=${PROJECT_ID} \
     --tpu-only \
     --tf-version=1.15.5 \
     --tpu-size=v2-32 \
     --name=amoebanet-tutorial-pod \
     --zone=europe-west4-a
   

3. Update the MODEL_BUCKET directory to store the Pod training data.

   (vm)$ export MODEL_DIR=${STORAGE_BUCKET}/amoebanet-tutorial-pod
   (vm)$ export TPU_NAME=amoebanet-tutorial-pod
   

4. Train the model on your Compute Engine VM with the following flags:

   (vm)$ python3 amoeba_net.py \
      --tpu=${TPU_NAME} \
      --data_dir=${DATA_DIR} \
      --model_dir=${MODEL_DIR} \
      --num_cells=6 \
      --image_size=224 \
      --num_epochs=1 \
      --train_batch_size=4096 \
      --eval_batch_size=1000 \
      --lr=10.24 \
      --lr_decay_value=0.88 \
      --num_shards=32 \
      --lr_warmup_epochs=0.35 \
      --mode=train \
      --iterations_per_loop=1000
   

   Parameter	Description
   tpu	Specifies the name of the Cloud TPU. Note that ctpu passes this name to the Compute Engine VM as an environment variable (TPU_NAME).
   data_dir	Specifies the Cloud Storage path for training input. It is set to the fake_imagenet dataset in this example.
   model_dir	Specifies the directory where checkpoints and summaries are stored during model training. If the folder is missing, the program creates one. When using a Cloud TPU, the `model_dir` must be a Cloud Storage path (`gs://...`). You can reuse an existing folder to load current checkpoint data and to store additional checkpoints as long as the previous checkpoints were created using TPU of the same size and TensorFlow version.

The procedure trains AmoebaNet-D model on the fake_imagnet dataset to 1 epoch. This training takes approximately 10 minutes on a v2-32 Cloud TPU. It will train to convergence in approximately 35 epochs.

Evaluating the model

Evaluation must be run on a single Cloud TPU device so you need to delete the Pod TPU and create a new single Cloud TPU device. You then use this Cloud TPU to evaluate the above trained model against the fake_imagenet validation data.

1. Delete the Cloud TPU resource you created for training the model on a single device.

   (vm)$ ctpu delete --project=${PROJECT_ID} \
    --tpu-only \
    --name=amoebanet-tutorial-pod \
    --zone=europe-west4-a

2. Run the ctpu up command, using the tpu-size parameter to specify the Cloud TPU you want to use. For example, the following command uses a v3-8 TPU type.

   (vm)$ ctpu up --project=${PROJECT_ID} \
     --tpu-only \
     --tf-version=1.15.5 \
     --tpu-size=v3-8 \
     --name=amoebanet-tutorial-eval \
     --zone=europe-west4-a
   

3. Update the MODEL_BUCKET directory to point to the Pod training data.

   (vm)$ export MODEL_DIR=${STORAGE_BUCKET}/amoebanet-tutorial-pod
   (vm)$ export TPU_NAME=amoebanet-tutorial-eval
   

4. Run the model evaluation with the following flags:

   (vm)$ python3 amoeba_net.py \
    --tpu=${TPU_NAME} \
    --data_dir=${DATA_DIR} \
    --model_dir=${MODEL_DIR} \
    --num_cells=6 \
    --image_size=224 \
    --train_batch_size=4096 \
    --eval_batch_size=1000 \
    --mode=eval \
    --iterations_per_loop=1000 \
    --eval_timeout=150
   

This generates output similar to the following:

Evaluation results: {'loss': 6.908725, 'top_1_accuracy': 0.001, 'global_step': 10955, 'top_5_accuracy': 0.005}

Since the training and evaluation was done on the fake_imagenet dataset, the output results do not reflect actual output that would appear if the training and evaluation was performed on a real dataset.

Note: In the evaluation command above, eval_timeout is set to 150, which terminates evaluation after 150 seconds. This is required because the evaluation loop is designed to run in parallel with training, where it keeps checking for new checkpoints for evaluation. The script may appear to hang if the evaluation loop is waiting for a new checkpoint to be written. By default, the evaluation loop stops after 48 epochs or 24 hours whichever is earlier.

Using larger Cloud TPU Pod slices

You can also try the model using larger Cloud TPU Pod slices. The following table shows the recommended values for these slices.

Caution: Scaling to larger Pod slices are experimental with this model.

Parameter	Description	128 cores	256 cores
num_cells	Total number of cells	6	6
image_size	Size of image, assuming image height and width	224	224
num_epochs	Number of steps use for training	60	90
train_batch_size	Global (not per-shard) batch size for training	8192	16384
lr	Learning rate	20.48	40.96
lr_decay_value	Exponential decay rate used in learning rate adjustment	0.91	0.94
num_shards	Number of shards (TPU cores)	128	256

Clean up

To avoid incurring charges to your Google Cloud account for the resources used in this tutorial, either delete the project that contains the resources, or keep the project and delete the individual resources.

1. Delete your Cloud TPU:

   $ ctpu delete --project=${PROJECT_ID} \
     --tpu-only \
     --name=amoebanet-tutorial-eval \
     --zone=europe-west4-a
   

2. Disconnect from the Compute Engine instance, if you have not already done so:

   (vm)$ exit
   

   Your prompt should now be username@projectname, showing you are in the Cloud Shell.

3. In your Cloud Shell, run ctpu delete with the --zone flag you used when you set up the Cloud TPU to delete your Compute Engine VM and your Cloud TPU:

   $ ctpu delete --project=${PROJECT_ID} \
     --vm-only \
     --name=amoebanet-tutorial \
     --zone=europe-west4-a
   

   Important: If you set the TPU resources name when you ran ctpu up, you must specify that name with the --name flag when you run ctpu delete in order to shut down your TPU resources.

4. Run ctpu status to make sure you have no instances allocated to avoid unnecessary charges for TPU usage. The deletion might take several minutes. A response like the one below indicates there are no more allocated instances:

   $ ctpu status --project=${PROJECT_ID} \
     --zone=europe-west4-a
   

   2018/04/28 16:16:23 WARNING: Setting zone to "europe-west4-a"
   No instances currently exist.
           Compute Engine VM:     --
           Cloud TPU:             --
   

5. Run gsutil as shown, replacing bucket-name with the name of the Cloud Storage bucket you created for this tutorial:

   $ gsutil rm -r gs://bucket-name
   

Note: For free storage limits and other pricing information, see the Cloud Storage pricing guide.

What's next

The TensorFlow Cloud TPU tutorials generally train the model using a sample dataset. The results of this training are not usable for inference. To use a model for inference, you can train the data on a publicly available dataset or your own data set. TensorFlow models trained on Cloud TPUs generally require datasets to be in TFRecord format.

You can use the dataset conversion tool sample to convert an image classification dataset into TFRecord format. If you are not using an image classification model, you will have to convert your dataset to TFRecord format yourself. For more information, see TFRecord and tf.Example.

Hyperparameter tuning

To improve the model's performance with your dataset, you can tune the model's hyperparameters. You can find information about hyperparameters common to all TPU supported models on GitHub. Information about model-specific hyperparameters can be found in the source code for each model. For more information on hyperparameter tuning, see Overview of hyperparameter tuning, Using the Hyperparameter tuning service, and Tune hyperparameters.

Inference

Once you have trained your model you can use it for inference (also called prediction). AI Platform is a cloud-based solution for developing, training, and deploying machine learning models. Once a model is deployed, you can use the AI Platform Prediction service.

• Learn more about ctpu, including how to install it on a local machine.
• Explore the TPU tools in TensorBoard.